Mikio L. Braun

ML and AI expert

@mikiobraun

@mikiobraun Twitter Memorial

19,827 tweets · 2008–2024 · 1046 threads

2023

Replying to @francoisfleuret
Is that what those AI based image enhancers are trying to hide from us?!
1
Replying to @mtantawy
Yeah, a bit of Twitter, now on threads, mastodon more passively via fedistats to be honest. Oh and Linkedin 🙈 Yeah, finding people and bootstrapping an account is really hard. That we still need to do this manually in 2023... almost 2024 🙄
1 View thread →
Replying to @_deep_yearning
You guys get out of your bed in less than 15 minutes after waking up?! Not to speak of getting to the desk... 😅
1
Replying to @Quasilocal
At this point maybe a movie style credit roll would make more sense?
3
Replying to @francoisfleuret
At least they could let an AI summarize those papers! What year is it? Summer of 2022?!?!
1
Replying to @francoisfleuret
The way I learned it, ℕ is {1,2,3,...} and if you want zero in it, you write ℕ_0.
Replying to @rorcde
Maybe you need to post more often! Joking aside, bootstrapping your followers on a new network is hard and tedious...
1
Replying to @CFDevelop
It was one of those features that sounded great but in reality these exception declarations would become so annoying to maintain that people rather dealt with whatever comes up the call stack.
1
Replying to @ChappellTracker
Thinking around the triangle is the new thinking outside of the box?!
1
Replying to @leonpalafox
Oh yeah, IMHO it is indispensable to really work through papers, especially when it's some approach you're just learning about. It you already have that knowledge, you can start looking for the main idea, but at some point you need to really understand it. Or at least that was me
Replying to @leonpalafox
Even worse, reading lots of papers and figuring out all the holes in the derivations yourself. With problem sets, someone at least tried to make them educational.
First impressions from Threads: I didn't know I was missing titles on link cards this much.
Replying to @ChappellTracker
I don't know how often I purchased Final Fantasy V and I'm too afraid to find out.
2
Replying to @francoisfleuret
Ha, amazing, I realized at some point that this is the same Björn Ommer who did his diploma thesis in the lab where I did my Ph.D. back in the early 2000s. He already was interested in vision and composability back then.
1
Replying to @dosinga
@Spotify Everyone knows December is just unhinged and that data can't be trusted anyway!
2
Replying to @francoisfleuret
Dropping out of the plane on a Monday evening directly into a grueling 5 hour poster session - I don't miss those days! 😅
Replying to @francoisfleuret
"Harsh reminder"? "The cold hand of regret reaching out from the past"? "A wake-up call to get your life back on track"?
2
Replying to @francoisfleuret
It's also not the first time, remember that demo where the AI called a hair saloon to book an appointment?
7
Replying to @mtantawy
Tell me we're attributing too many human qualities to software systems without telling me we're attributing...
1
Replying to @tunguz
True story, I had to tell the young ones about this. They had never heard of it!
RT @francoisfleuret: The number of qbits is totally meaningless. There is no other proper measure? An entropy or something?
Replying to @mariofusco
The key for me is to listen to music that I already know very well, then it can be anything, even something that is labelled "distracting" here.
Replying to @alper
Man, who let them fit quadratic functions to this data?!? But yeah, my general impression is that so many places (Behörden, Deutsche Bahn, health...) in Germany are on the brink of collapse but somehow haven't completely crashed yet.
1
Replying to @rorcde
Yeah, there's a reason while sama is walking around talking governments into expensive regulation, he's trying everything to put up moats...
1
Replying to @rorcde
The number of times I was in a meeting where a manager asked "how long would this take if we add some people to help?" is unfortunately non-zero.
1
Replying to @rorcde
Same same :) Internal infrastructure got a bit more streamlined which is good, but internal tooling vs. tooling you can buy is always a stretch. Documentation & onboarding experience is much more polished for external tools.
1 View thread →
Replying to @rorcde
Yeah, I'm not sure especially about Apple and being concerned about OOS and "non-aligned" AI solutions. They also have been adding some ML functionality to their APIs for years now, but I see nobody talking about that.
RT @rorcde: Local LLMs running on devices are where we are headed. Like transitioning from mainframes to handheld devices, we will experi…
Replying to @rorcde
Yes and I also believe we'll find models that require a lot loss computing power. As you said, that's the way of technological progress. First you create the new technology, then you optimize and streamline.
1
So because of Alan Wake 2 I've been replaying other Remedy games like Quantum Break. This scene with the chopper frozen in time was still in my memory, but I forgot that it was almost at the end. It had always reminded me of this photograph by Andreas Feininger.
Media Media
3 View thread →
Replying to @rorcde
Yeah, I was wondering. At the time it felt like a bit too good to be true. Providing the description by plain text and then it magically figures out when and how to call your API? I thought maybe this was the first draft and they got it to work reliably, but apparently not.
Replying to @krishnanrohit
It's really hard to read this in an office and not to be ROFLing.
RT @wgluckin: Take notes, growth hackers: my German insurance company just sent me a physical paper letter to tell me about a new app featu…
RT @fchollet: I'm pretty excited about the "deep learning" thing, but I'm also worried about being excited about a thing around which ther…
Replying to @mtantawy
Yeah, I had that, too. I barely made it in time, but still saw my appointment number on the board, then went to the room, and then the guy said "oh, I just took that number off the board because nobody showed up". But then he still processed my case, so there's that 😓
Replying to @mtantawy
The only thing that ever works for me is to look at the page between 7 and 8am and then reload it till miraculously new slots for the same day pop up.
View thread →
Replying to @noootsab
@DB_Bahn Hey man, yeah, this was a quick in and out (at least that was how it was planned), I'll let you know! And yeah, luckily I had a Switch and also bought a replacement power adapter for my Mac (which I had forgotten)...
I'm not going to link to this but how an experienced scientists makes disagreement about someone else being selfish to weigh his own opinion more. Okay, I'm reading the "being selfish" into this, but I thought the whole point of science is to go beyond opinion.
2
Thread (6/7)
Even in the middle of the night, the DB can still add more delay, new ETA for Berlin is 2:02am. I've had a couple of unlucky trips, but I think this is a new record.
Media
Thread (5/7)
To anyone who is still following, we just started moving again. New ETA for Berlin is 1:45am - original plan was 10:38pm - more than 3h delayed on a trip that was supposed to take 4.5h.
Replying to @SashaMTL
So much of today's "news" seems to be to go in with completely false assumptions and then be outraged at what happens.
Thread (4/4)
@bocytko They just explained that we really have to wait here till the track is cleared because possible detour routes are also blocked or under construction. Excellent.
Thread (2/4)
@bocytko I know that they have people working on better planning and "AI" and stuff, but the time horizon for this seems to be "everything will be fixed in 30 years." And then there are minor improvements like real-time updates in the app, communication transparency is much better, etc.
Replying to @bocytko
Yeah, obviously I'm no expert, but in some areas, the whole system seems always on the brink of failure. There's no resilience, small delays amplify through the network. The planning is also not adapted, realistically you should always plan in 30 minutes for connections.
View thread →
Thread (4/7)
A week ago I was late more than 2h on a train to Munich. Kudos to the DB that they processed my compensation within days. I got half of the fare back on Tuesday. Still, this just means they lose more money, this is clearly not sustainable.
2
Thread (3/7)
Deutsche Bahn is falling apart. Especially NRW is just an unpredictable mess. I don't think this is complaining on a high level anymore, it really is pretty bad.
2
Once again stuck in a @DB_Bahn train from NRW to Berlin. This one left Cologne already 30 minutes late, then random delays, now another accident along the line, projected delay is 2h.
2 View thread →
Thread (2/3)
On a more pedantic note, they define frontier AI as foundational technology that could become dangerous. Not sure if I read it right, but the very definition contains that assumption it could become dangerous. Sounds pretty circular to me.
I read that EA AI policy paper that's making rounds, and let me say that, in general, going through life defending yourself against all those things that /could/ go wrong is a bad strategic guaranteed to make you miserable.
3 View thread →
I forgot my charger and didn't really paid attention and bought the 35€ Apple vinyl mesh 2m USB-C cable, AMA.
Replying to @wesbos
What, this only works within the same process? Then again, these windows need to be on the same screen so yeah...
This is *not* a subtweet, but when the 10x engineer is the one who cuts through all the red tape and delivers value in spite of all the processes that are in place to provide "guardrails" that's a huge 🚩🚩🚩 for your company.
7
Replying to @leonpalafox
@zaxtax We're already following each other, if you want to talk more, send me a DM!
2
RT @RemivanTrijp: LLMs don’t hallucinate but generate. Sometimes their generated output is correct, sometimes not. Calling the latter hallu…
RT @zachreinert0: Elon has lost his wife, his kids, 40 billion dollars, and his space ship crashed. It’s like a genre of country music that…
Replying to @vivekhaldar
Yeah, but learn time and again that there are many people who are not on social media. Given that I think the news sites are doing an ok job.
Replying to @noootsab
No, I'm not even sure what I wanted to demonstrate, I just had a pretty graph and needed some text 😅
RT @haro_ca_: Data Scientist with his MacBook Pro M3 Max signing in to databricks and running queries on a cluster https://t.co/bd37cibxXY
Media
Replying to @francoisfleuret
Would you recommend the RTX 4060Ti? I see that it's the lowest tier that supports 16 GB RAM. Is that the main reason for it (and it draws half as much power as the 4080)?
Replying to @krishnanrohit
This was 5 hours ago, in the meantime, there's a new board and Sam is coming back. Yay... sort of... Somehow I feel like all of this shouldn't have been discussed so publicly.
1
Replying to @ridtalkstech
Yeah I think once they saw the dollars rolling in they threw out almost everything about the original idea of OpenAI...
Replying to @bocytko
@fedistats Oh I'm just putting them in a postgres tsvector field - doesn't make any sense for base64 encoded pictures but who knows... 😅
Thread (2/5)
@fedistats Sooo, what happened is not that it crashed up that someone put a base64 encoded image into their og_description tag 🤯 People, always sanitize and check data you get from random servers on the Internet!
4
RT @josh_wills: Can’t wait for all of you clowns to go to sleep so I can thought leader the shit out of this openai thing from Taipei
Replying to @mtantawy
Well at least that won't take a lot of money to build... please tell me they didn't get millions to build this!! 🥺
Back in the 90s when I started to study CS some (non-tech) people asked me why I went into this field as "computers would just program themselves anytime soon." I laughed them off for decades, but maybe... the time has finally come?! 😅 twitter.com/simonw/status/…
6
Replying to @francoisfleuret
@EmilStenstrom I knew I should've gone for the 64GB Macbook Pro. But no, I thought at the time, that would be insane 😅
1
RT @arthurmensch: We have heard many extrapolations of Mistral AI’s position on the AI Act, so I’ll clarify. In its early form, the AI Act…
RT @pmarca: I am calling for a total and complete shutdown of AI Doomerism until we can figure out what is going on.
RT @fedistats: Tired of opening Twitter erm X to stay up to date on the latest AI news? Kind reminder that you can get a daily email with…
Replying to @CreeCoder
@TheJackForge No joke, the Apple Silicon MacBook Airs are more power than you need for most things. The only thing that really bugs me is that you can only have one external monitor.
2
Replying to @bocytko
@krishnanrohit The idea was to get rid of JSON, not enforce it!! 😋
1
Oh, did Melon remove the "Not interested in this ad?" option. Well... I guess it's time for blocking then...
RT @jayjen_x: A quick, no BS recap of OpenAI Dev Day announcements So you don't have to watch a 1 hour long live-stream: GPT 4-Turbo • GP…
Replying to @krishnanrohit
Yeah we've been under the yoke of things like JSON for way too long!!
1
My son asked me what a meeting is. "So basically 10 people in a room and you sit around and get paid?" 😅
1 7
Replying to @denis_bvk
@borovikov_en My thoughts exactly. I think this illustrates how much non-tech people's understanding of tech is lacking.
2
RT @manekinekko: Me getting a shiny new mech keyboard be like #keebs https://t.co/LWbp5Fg4j6
Replying to @paul_rietschka
I think it perfectly captures our inability to measure weight well. Adding all those irregular conversions ratios is another story.
Replying to @paul_rietschka
Yeah, I went for 18g because that's what I read on the Internet. That's for one cup of coffee. I feel like it's more consistent now. OTOH different beans lead to widely different volume of ground coffee. No idea what's up with that 🤷🏻‍♂️
1
Still thinking about that one senior dev who went from apps to backend and then said backend is not for him because it's so intangible.
Replying to @mrclbschff
@flaviocopes This! No joke, I bought a M1 Max with 32GB and later regretted not just maxing out the RAM.
Replying to @rorcde
@mtantawy I'm learning more about Windows and ARM today than on any day before basically.
1
Replying to @InkmiHq
@rorcde Ah, and yeah, I think the article could be interpreted that Intel also still sees AMD as its main rival, not ARM (and ARM is just the IP, not even a specific manufacturer).
Replying to @rorcde
@mtantawy Interesting! I don't know what Apple does, but it seems to be quite good :) I remember when they switched from PowerPC to Intel they also had this technology but I think it wasn't as good back then (haven't used it myself, though)
2
Replying to @InkmiHq
Yes, they're not. What I meant is that you can pay the premium and still get "non premium" CPUs
Replying to @rorcde
@mtantawy So you can run it on ARM, but just with ARM compiled binaries? I don't know how Apple pulled it off, but being able to run Intel binaries on ARM at about the same performance was also a killer feature. Microsoft should have the expertise in house to build that as well... .
Thread (2/2)
@InkmiHq Regarding luxury brands, I had the impression that the past few years Sony and the like also started to build sleek looking laptops at similar price points. Which is fine, just that in terms of CPU tech they are lagging so far behind now... .
1
Replying to @InkmiHq
I'd personally love to see AMD/Intel get better on the power side. I personally don't have anything against the architecture, I think being more extensible is also good (especially for desktop). With Apple Silicon you have to decide how much RAM/which GPU to get when you buy it.
View thread →
Replying to @InkmiHq
OK, with world apart I mean that the M1 has 4x the cores of the Intel, and the Intel frequently heats up and the fans are running at high speed while the M1 does not even have a fan. So something like 4x faster at half the power, is that roughly 10x? :)
Replying to @rorcde
@InkmiHq Oh yes, LinkedIn is the worst. On my Intel MBA it was barely usable. Typing in the chat was limited to 2 cps.
Replying to @InkmiHq
M3 Ultra GPU performance could be great. I'm looking forward to first benchmarks of the ray tracing performance... .
1
Replying to @untitled01ipynb
At this point this app should start feeling like some stuff out of Night City anno 2075, instead we're stuck with San Francisco, 2011.
1
Replying to @untitled01ipynb
NFTs even that eventually become fintokens on the X superapp!!
Thread (2/4)
I still haven't made the math but isn't any cloud data center immediately above the threshold? Does the US administration know how cloud computing works?!
Replying to @mtantawy
As soon as Microsoft starts supporting ARM fully it is over. Anyone who still believes that ARM is just good for smartphones has never used an Apple Silicon computer.
1 View thread →
Replying to @InkmiHq
Yes, although I feel the practical difference is very real. I have both an Intel and an M1 Macbook Air and just in terms of performance and heat (= fan) generation, they are a world apart.
Replying to @paul_rietschka
Oof, not a good combination! And yeah, the cold is not the problem but the lack of sunlight in Berlin. We regularly go through weeks of constant overcast sky and no direct sunlight.
Wait people are surprised that OpenAI can now inject PDFs, too? So (1) have they never heard of the term "sherlocked" and (2) do they know how easy it technically is to extract text from PDFs? Use something like Apache Tika and you have a 0th iteration.
3
Oof, first working day after the end of DST. It's half past four and it is already dark in Berlin. #WinterIsComing
2
Replying to @mtantawy
The other day I learned you can buy a postage stamp which is just a code you write on the envelope. 🤯 if that actually works
1
Replying to @FlaxSearch
I'm 100% with you, Charlie! If this is band aid that's being put over horrible document writing requirements in corporate environments, it needs to be ripped off immediately!
RT @FlaxSearch: I don't need or want to 'have a chat with my PDFs', I'd rather have a chat with the people who wrote them. It's a horrible…
Replying to @rorcde
I think the average Berlin restaurant not accepting electronic payment is not a infrastructure problem, more a tax evasion problem 😅 And I heartedly disagree with the Elon the Savior narrative, but that's a discussion for another day... .
1
Replying to @rorcde
To be honest I think there's a world between Sparkasse and whatever Musk is doing... just because companies also undergo mergers and bankruptcies doesn't mean something that's so unhinged like 🍉 can also be trusted.
Replying to @fmueller_bln
@InkmiHq I don't know whether that's the most cynical or the most ingenious idea around AI I've heard in a long time 😂 Yeah, you know how people respond to social proof, but what if the social part is all AI?!
1
Replying to @rorcde
Maybe I'm too German for that, but if I put all my financial life into a company, I want stability and to know it will still be there 50 years from now.
Replying to @rorcde
Well, yeah if they manage to get that... let's see. What I've seen is very erratic product decisions, laying off teams that are about security, treating customers as bad children, being in love with a tech vision that has no validation... nah...
RT @shanselman: Shift-Ctrl-T reopens the tab you just closed. It's like undo for closed tabs and you can repeat it over and over. Get this…
Replying to @rorcde
Regarding Bluesky, I don't know, man. Seems like the right folls aren't there yet or I haven't found them yet. Bootstrapping a social network is quite exhausting...
1
Replying to @rorcde
I fear that's really because being back to a fulltime job I spend less time on social media? 😅
1
Replying to @rorcde
Ha :) I made a shortcut to open this app and gave it the old Twitter logo. Downside (?) is I don't see the notification count bubble anymore.
Again, I still have invite codes for non-rainy skies. Let me know if you're interested to check out the grass on the other side.
3
Replying to @paul_rietschka
No need to be sorry, once I realized who the last two are I lol'd :D
1
Thread (2/4)
@paul_rietschka Ad algorithms are also an absolute joke. Especially these days I'm seeing ads next to the most horrific content. Someone is asleep at the wheel.
1
Replying to @paul_rietschka
I'm running out of snarky comments. IIIRC, solving the bot problem was one of the claims that were made. And yet, a year later we're at the same spot with respect to that, and so much worse in almost everything else.
1 View thread →
Replying to @mtantawy
@ahmdelemam Deleted my account years ago. But I'm still using Instagram, so there's that 😅 But yeah, I don't know whether that's a real quote, but I once read that Zuck said something like "there's a way to be unethical and not break the law and that's how I want to live" and it fits.
Replying to @rorcde
Yeah, and then later they praise themselves for all the great new features they have been pushing...
1
Replying to @rorcde
Yeah, I think you said that a while back but I agree it's even more true now.
1
Replying to @rorcde
Yeah, it's super confusing but I read he likes it because it is less cluttered. MElon is really every technical founder ever who just implements his own preferences instead of doing proper customer research...
Replying to @rorcde
@paul_rietschka Of course! I mean if the data allows it :) Compared to trees, the main shortcomings are that they cannot deal with nominal data out of the box and that you sometimes need extensive cross validation for hyperparameter tuning.
Replying to @paul_rietschka
Ah, it actually is called "the OA." I thought it might be an acronym :)
1
RT @wrede: I am going to be using Twitter a lot less going forward. Please join me here instead https://t.co/FMb8heBYP9
Media
Replying to @paul_rietschka
The one where someone argue with him his name cannot be pronounced that way is somewhat relatable... the rest, yeah well.
1
Replying to @paul_rietschka
Yeah, Yann was special in that (1) he never strayed from NNs (2) he was always putting together nifty demos even at levels where others became pure science managers and (3) he has a pretty funny website.
Replying to @paul_rietschka
I have not but yeah, joking aside I'd also start (and most likely end) with xgboost.
Replying to @paul_rietschka
Sometimes I look at the soft margin optimization problem and mutter "ah well, the young ones don't know what they're missing out on."
1 View thread →
RT @FlaxSearch: As it's Friday it's (hopefully) humorous analogy time. I started my career as an engineer and I love building and fixing th…
Replying to @InkmiHq
@KingOfCoders Oh yes, it is definitely very helpful, but what I meant is that that's not the main challenge for teams in big orgs.
Replying to @paul_rietschka
Yeah I'm reconsidering whether I became the right kind of architect...
1
Replying to @paul_rietschka
Making it and then designing and building your own home seems to be the architect's equivalent of "I'm running my blog on my own framework from the raspberry pi in my living room."
Replying to @mleznik
Yeah, if they find something that can reliably refactor mountains of legacy code, I'd be impressed. And then there are all those non-code issues like defining team responsibilities, interfaces, aligning roadmaps and architectures on different levels, politics, etc. etc.
1 View thread →
Replying to @terronk
Is the idea really that people click on more stuff because they want to see what it is about? I predict og:image will embed the text soon if they are asked by Twitter.
1
People who think that chatGPT & friends will significantly improve developer productivity never faced the challenge of dealing with the complexity of writing software in big organizations.
9
Replying to @dirkriehle
I know right?! I realize there is so much I need to do before I can leave the house!!
1
Replying to @rorcde
100%! 10k is very solid. I think during the lockdown in winter when I stayed inside, I probably only did like 1000 steps or so. Our bodies were not made for so much rest!
1
Thread (2/2)
@rorcde And chat, yeah, my experience is you either need to be very formal or you need to know each other well. And even then there are misunderstandings sometimes...
Replying to @rorcde
Yeah, I also find myself coming back to good old phone calls over chat or video. I think video has more bandwidth in principle, but the lag is much more obvious and that makes it less natural. Phone calls have that sometimes, too, but it isn't that obvious.
View thread →
Thread (2/2)
@rorcde And then - and I never thought I say that - my body just starts aching after a few days of pacing around in my flat. Just commuting to the office is already some well needed workout 😅
Replying to @rorcde
Yeah, not every day, I also have a good setup at home, but over time I realized I need those human interactions, too. I think there's also a lot of trust and relationship building happening around meetings that are easily missing from virtual meetings.
1 View thread →
Replying to @rorcde
Hybrid. TBH I forgot what the ratio is, I was just excited to get out of my little home office again 😂
Replying to @rorcde
Unpacking equipment and welcome pack. And watching the reactions I got on LinkedIn. 😊
1
Replying to @rorcde
Let's see :) It's the part that takes care of orders, so much I can say.
1
Replying to @rorcde
Yeah, less time for fedistats 😢 Yeah, I think many companies laid off people, and then there was no budget for freelancers and consultants as well. It's slowly coming back now I think...
1
Man, posts on starting new positions always get so much engagement on LinkedIn, almost makes you want to switch jobs more often... for the clicks. 😅
7
Replying to @InkmiHq
@KingOfCoders Yeah, same here. I always felt that the "marketing" about rust gets it wrong. It's a nice language with proper generics, strong typing system, etc, and yet they make it all about memory safety. I don't lose sleep about memory safety (*as long as it sort of works).
1
Replying to @shanselman
What I find amazing about game pass is that I've read some studios also really prefer this model as it takes out some of the uncertainty about whether your game will be an economic success or not.
3
RT @wrede: Ah, Product Hunt. You have disappointed me so many times. It's probably always been my fault, not yours. I'm going to give you…
Thread (2/2)
@paul_rietschka That's a good Wikipedia article. Also, by its very nature, I'd consider "disruptive entrepreneurship" to be really hard to predict...
Media
1
RT @jimprosser: When you don't have a comms team in place to sweat the details, you end up pictured in a crucifixion pose for your big FT p…
Replying to @chrisalbon
I think one only needs to see who is pushing for AI regulation... Altman and friends have a very real business interest in this.
1 12
RT @tunguz: Great news for all IoT and Edge Computing enthusiasts and all others who value small capable computers: the Raspberry Pi Founda…
Replying to @rorcde
Yeah yeah, last I heard was he wanted to introduce a small monthly usage fee because this was "the only way to get rid of the bot problem."
1
Replying to @rorcde
I never used it but from time to time saw a tweet that had the notes with it and usually it was informative. But it was also stuff like an add for a mobile game and the note said that the game looked nothing like this. Still useful!
Replying to @paul_rietschka
Yeah, no shade on Ive. Especially since he got rid of the skeuomorphism.
1
Replying to @paul_rietschka
Yeah, it really got worse with each person mentioned. Jony Ive... ok... Sam Altman... ah, not thanks... Masayoshi Son... OMG! O_o
1
Replying to @francoisfleuret
The "best" thing that once you wrote too much text, you started overwrite the beginning of your program O_o
1
Replying to @mariofusco
I don't know, it's the sign of the times, people focus on easily digestable bits of information. Talking about anything more complicated seems to be too much. At least on social media... .
1
Zweite Reise mit dem Sprinter Berlin - Köln mit der @DB_Bahn. Gestern +50min, heute nur +10min bis Berlin, aber dafür dann Anschluss nach Südkreuz nicht bekommen, nächster IC wartet 30 min weil Lokführer nicht da, wieder +50min. Läuft.
1 2 View thread →
Replying to @tunguz
It will not just change the game, it will put the concept of game changing on an entirely new level.
2
Replying to @untitled01ipynb
Ok, that made it infinitesimally clearer, but I am still here to see.
1
Replying to @untitled01ipynb
I don't even know what a roona infra team member actually is but here have my like I want to find out!
1
Replying to @mtantawy
Yeah, it took a bit longer for me, but I eventually got there. The most absurd version of this is where "becoming successful" means having a lot of followers and how to make money off the advice how to achieve this. It is very circular.
1
Thread (2/2)
There are some things that are open to all. Like learning a skill and finding a job. Then there are these once in a million chances. Sure, you can try and work towards it, but it's not like it's a repeatable process. Just imagine all their 50k followers would just pull it off.
1
That's it, I'm unfollowing all the "I got incredibly successful / you can do it too / this is my playbook" kind of accounts. I used to find this interesting and somewhat inspiring, but in reality, these are mostly pipedreams.
6 View thread →
I heard there's demand for bluesky invites? Let me know if you're interested, I still have a few codes.
1
Replying to @krishnanrohit
@nabeelqu I keep telling the young folks about winamp. They have no idea. The endless possibilities!
Replying to @pdrmnvd
That is A LOT of data. Probably more than a freight train full of CDs could hold.
1
Replying to @Major_Grooves
@DB_Bahn Yeahyeah. I believe 5 minutes is even officially "not delayed."
1
Replying to @alung
@KingOfCoders You got a notification from Google about this? At one of your... ahem... former employers, or you privately :)
Replying to @InkmiHq
@KingOfCoders Why do they want it anyway? I thought you need a reason to collect it (by GDPR? Apple app store rules?) But if they adjusted the font that would even justify collecting the birthdate! 🤔
Replying to @leonpalafox
Yeah I still remember people were really looking down even on Java because it wasn’t directly compiled to machine code… A lot has changed, machines got a lot faster but people also figured out which parts you need to optimize.
Replying to @leonpalafox
I tried to build tf a few years back because my processor was so old it didn‘t support some of the vector instructions. I gave up after 7 hours or so. I‘m pretty sure it tried to bootstrap gcc somewhere inbetween. Maybe they streamlined it, but that told me a lot about tf.
RT @paul_rietschka: The resurrection of multibackend Keras is probably the most impactful software development we’ll see in the data world…
RT @alung: Hire me at your data/music company so I can ship a clashfinder resolver during hackweek
Thread (2/2)
It's a bit I don't trust that the C compiler handles pass-by-value for struct because the first compilers I used didn't support it. There is no guarantee of delivery for emails! So much can go wrong!! Spam filters, misspellings, servers rejecting emails, ...
@astrobassball I got three left myself. It seems they‘re not as popular as they used to be…
1
Replying to @paul_rietschka
Ha totally forgot that he also wanted to do „real“ money.
1
Replying to @tunguz
Lemme guess, you have to sign up for X Blue to know whether you‘re eligible?
Replying to @WangUWS
@paul_rietschka My guess it is either „none“ or „too many to make sense of them“
Replying to @alung
Yeah, the GTX 1660S was my last purchase to do some small scale deep learning, but now it‘s my son‘s main machine :) Gamingwise I‘ve moved to consoles, much better experience.
Replying to @InkmiHq
@KingOfCoders Hehe, I guess Apple‘s margins are big enough… don‘t they sit on a huge pile of cash?
1
Replying to @alung
AGP? It has been a while! 😅 Yeah, probably also „good enough“ for now. Although I think they iterated with wider PCIe like x16.
Replying to @rorcde
No, at home it‘s an (equally quiet) MBP 16“… Technically you can also take it on the road, and I tried it but practically it is just too huge and heavy 😅
1
Replying to @InkmiHq
@KingOfCoders Aha, interesting thought, you mean so they can run all their devices on the same hardware platform?
Replying to @rorcde
Well, I *really* like the „old“ Macbook Air drop shaped form factor. And this is the „on the go“ laptop, so I figured I don‘t need the latest cutting edge performance. Plus you can get them cheaper.
Replying to @Bediko
Hehe, oh ja. Dafür läuft es jetzt ganz stabil :) Aber eine sehr berechtigte Frage. Es hat echt lange gebraucht, bis mir klar wurde, was da eigentlich alles drin ist. Ist schon alles ein bisschen komplex im fediverse 😅
1
Replying to @Bediko
Es gibt Leute, die ziehen mehrere public timelines und merge die und so, aber das erschien mir für einen Prototypen zu viel Arbeit 😅
Replying to @Bediko
Äh also ich ziehe den public stream von einer Instanz und da ist alles drin, was die zu sehen bekommt. D.h. schon global, aber doch biased.
1
Replying to @paul_rietschka
The illusion of transparency and data driven decision making would be my guess.
1
Thread (2/3)
Mind you, this is very different from the 2000s where I often had to buy a new GPU to play the latest games. Which often meant buying a new mainboard because the expansion slot format changed. Anyone remember VESA local?
Replying to @paul_rietschka
I think we need to bring in "the AI" to help us make sense of all those dashboards!
3
Thread (3/4)
@KingOfCoders I always have to think about an article I read a while back how Apple is organised by areas of expertise and how they manage to make everything pull together nevertheless. Makes you think.
Replying to @mtantawy
Yeah, at the end of the day, it's not just the graphics but also the gameplay and how everything fits together. I'm still waiting for that insight to hit the AI space, there it's really just "bigger is better" for now.
1
Thread (2/4)
@KingOfCoders Apple definitely played the long game, first starting with chips for their iPhone to build up technology and expertise. Some criticised M1 for essentially being blown up mobile processors, but they are pretty impressive.
Replying to @InkmiHq
@KingOfCoders Unlimited money and also the willingness to move everything to a new hardware platform. Having full control over the entire product definitely helps. In comparison, Intel/AMD seem to be hindered by legacy.
View thread →
So I finally caved in and updated my MacBook Air dual core Intel to the one with M1 and it is every bit as amazing as I thought. 4 efficiency/4 performance cores is the best combination for this package. Still can't get over the fact that this thing doesn't have a fan.
1 View thread →
Replying to @mtantawy
Yeah, the one thing it really cannot do is raytracing, but then in every game I saw it, it felt a bit gimmicky and not really adding a lot. The best was probably Watchdog:Legions.
I have a GTX 1660S and it runs all games just fine. This 4 year old card is still good. Nvidia desperately needed something else to keep selling fridges and they found AI.
2 2 View thread →
Replying to @rorcde
Yeah, but there's been some buzz the past few days of people claiming that they understood the new algorithm. One was posts with a lot of engagement, so that people tried out question format. The other was very simple statements.
1 View thread →
Replying to @francoisfleuret
Also, wait for my upcoming book where I disclose 5 tips that helped me unlock my full potential and become a world renowned TED(x) speaker!
4
Replying to @paul_rietschka
Yeah, it's a pipe dream. Others have said that already, but I would also never trust Elon with my money.
Keep ignoring me but I‘m telling you, after video game graphics have leveled off, AI is the new killer app that drives GPU sales!! Nobody has an interest in smaller network!
1
RT @StephanEwen: 📢 We are excited to share a first look at our new project: Restate (@restatedev) 🔥 Building applications with 𝗱𝗶𝘀𝘁𝗿𝗶𝗯𝘂𝘁𝗲𝗱…
Replying to @paul_rietschka
People like simple explanations and tangible stuff. And OAI and friends are playing them just right, overemphasizing stuff like number of weights. I‘m sure there are historical examples of this, but probably not on this insane level of hype.
1
RT @julien_c: Personally, I don't subscribe to the Space-race type idea that the one company with the largest number of GPUs will reach AGI…
RT @mikiobraun: @paul_rietschka I wonder what other apps could be rolled into Twitter to make it „super“… - definitely a todo list - payme…
Replying to @paul_rietschka
I wonder what other apps could be rolled into Twitter to make it „super“… - definitely a todo list - payments obviously - MEME GENERATOR!!! - video clip editor - location sharing - captcha microtasks!! There has got to be more!!
1 2
Replying to @paul_rietschka
Elon… wants that X helps you with hiring??? Oh the irony!!!
1
Replying to @paul_rietschka
I absolutely agree. I think it already is not sustainable but kept alife by VC money.
1
Replying to @paul_rietschka
Weird, I keep thinking that‘s the most outrageous thing about this. But this is just like when I did approximate counting when everyone else was just scaling out… I‘m just old I guess. ;)
Replying to @paul_rietschka
Sooo what you‘re saying is now‘s the time to start shorting Nvidia…??!?
Replying to @paul_rietschka
Yeah, we definitely need the room-filling-super-computer to portable-compute-power-in-your-pocket crunch, and we need it soon. Although I'm not sure that will come entirely out of hardware improvements. But maybe that's what people in the 70s thought as well?
1
Replying to @boydroid
Find auch golden, dass man erst mal subscriber muss, um dann zu erfahren ob man überhaupt dabei ist oder nicht...
Replying to @InkmiHq
@KingOfCoders @NVIDIAGeForceDE Bin ehrlich auch eher skeptisch! Erstmal gucken was die anderen machen…
Replying to @InkmiHq
@KingOfCoders @NVIDIAGeForceDE Oh Mann… ja, alles nicht so einfach mit dem elektronischen Handel. Überhaupt, mal schauen ob sich dieses Internetz durchsetzt 🤦🏻‍♂️
1 2
Replying to @paul_rietschka
It never was… honestly, letting these things drive around is criminal neglect. What‘s the process for letting machinery on the street in the US? Is there any?
1
Replying to @mucio
@paul_rietschka Yeah, Linux is mostly entirely command line anyway… but you said something earlier about UI programs that don‘t integrate well. What were you thinking of?
1
Replying to @paul_rietschka
Yeah, I think some brands have tried to move into that market, but even their clean designs aren‘t as clean. And then you‘d still have Windows… The Mac minis are also beautiful devices, hard to rebuild. The CPU fan alone is already too big.
1
Replying to @paul_rietschka
Yeah, I never fully understood why nobody works on full hardware support. „Open source“ was a bit tainted in the past but I think we‘re way past this. Apple somehow managed to sell their OS as „close enough to *nix“ Windows has WSL which I haven‘t tries but is close, but yeah
Replying to @paul_rietschka
I‘m not really using Windows but my kids have it for gaming and stuff, and occasionally I sit down at it to run updates and it still is - in 2023 - an amazingly painful experience.
1
Replying to @fmueller_bln
Depends! For batch compute jobs I definitely want to see 100% per core!!!
1
Replying to @krishnanrohit
I once was witness to Yann and my prof having a little „who knows yachting spots in the Carribean better“ match over lunch… we PostDoc knew to keep out of it 😅
1
Replying to @mtantawy
The weird thing about the book is that it is written as a guide about how to write apps that engage people, but just like that article, you suddenly realize all the ways in which you are manipulated.
1
Replying to @untitled01ipynb
The funny thing is is that Meta isn‘t even in the business of selling AIaaS!!!
1
Oh great, first they make the tab bar disappear when you scroll, now they made it 30% translucent because maybe people have been complaining where it went… worst of both worlds.
Media
1
Replying to @paul_rietschka
Yeah, South American animals had their 15 minutes in the limelight! Time for something new!
1
Replying to @forgebitz
Yeah, when I was starting to study computer science in 1995, people were already asking me why, because „machines will be able to program themselves soon.“ It seems to be a deep rooted wish in the general population.
1
Replying to @John4man
How does this tweet only have 792 views?! All that‘s wrong with TSFKAT in one picture!!
1
RT @pdrmnvd: @John4man @mikiobraun i’m bringing back the hadoop stack if it don’t look like this i don’t want it https://t.co/dWrt7qz9jm
Media
Replying to @pdrmnvd
@John4man We should never have questioned the JVM!!! That‘s when Bad Things started to happen!!!!
1
Replying to @mucio
I was thinking more along the lines of spirit animal, but yours is good, too! :)
I just misread "sprint demo" as "spirit demo" and now I'm thinking what that would be like.
1
Replying to @CFDevelop
Yeah, they're essentially adding banners to the google search for chrome, the chrome homepage etc. Some of it seems to be paid but some also seems quite shady actually. The vibe is also very desperate.
1
Replying to @CFDevelop
Radical new idea: you build the DOM once and then serve that to the client. It's incredibly fast! Don't give me ideas!!!
1
Last week I did a clean install of Win 11 and the amount of messages when you try to install Chrome was staggering. My favorite one was a banner saying „you know Edge uses the same engine but is just… better?!“
1 2
Replying to @paul_rietschka
@tcarpenter216 I also got two left. #tryingtonotsounddesparateoranything
Replying to @paul_rietschka
My brain plus DuckDuckGo search. Yeah, I‘m oldschool/weird like that.
Replying to @ridtalkstech
Delete! If the pandemic has told me one thing it‘s that my body can‘t function at 300 steps per day.
Thread (2/4)
@francoisfleuret Which reminds me of people saying that before copy machines people needed to write summaries by hand, which really helped with learning, but now everyone just copies. What‘s next? People just talk to an AI explaining it to them and don‘t even read?
Replying to @paul_rietschka
@Kyle_Ireton Yeah, can‘t even get those invites out, that‘s how it is with bluesky.
2
Replying to @krishnanrohit
No one ever talks about this?!? Maybe… because it was known?!
1
Thread (2/2)
Another time he said "A student came to me and asked whether I could pass him already because he got an internship in New York. I said to him 'I'd also like to do an internship in New York! Sometimes in life you have to decide'." Invaluable life lessons! 😅
1
Thinking back to my algorithms professor back at university (Arnold Schönhage). At the beginning of the lecture he said "You know this lecture is called efficient algorithms. Don't copy your exercises, forge the course certificate!"
5 View thread →
Replying to @krishnanrohit
My impression from devs in their 20s talking on TikTok about their „total comp“ is that 150k is at the lower end?
10
RT @blprnt: One thing the 🟦 app is sorely missing is representation from critter twitter. If you post about strange creatures here (nudibr…
Well, with all that is happening, and not like a lot is going on at BlueSky, but I have 3 invite codes, FIFO!
Well, like clockwork, MElon put out another egragious tweet on why we even need blocking... as I said I think we need to move off big social media sites, and create protocols that will moving among sites much more frictionless.
Media
1 View thread →
Replying to @paul_rietschka
Yeah, I also always thought the media difference is a huge one, but it was just a hunch. It's probably much easier to get that dopamine hit with a short video than a thoughful/inciting/sarcastic tweet :)
1
Replying to @paul_rietschka
Yeah, so are the rumors true they forked this from the instagram codebase? That would explain why no desktop app (although the instagram website is reasonably close to the app). I don't get why no DMs though, don't they have that packaged up as a library somewhere on some repo?
Very happy with my „iOS automation that just starts another app on the homescreen with a custom icon.“ On the plus side, it naturally doesn‘t display badges, so a bit less FOMO :) I almost forgot what it is called now 😇
Media
2
Replying to @rahuldave
@untitled01ipynb 😂 Well, because you first need to clean the data! 🥁
Thread (3/5)
That way, even if your matrices are in the GBs, you can fully saturate whatever computing pipeline you have by working on peaces that fit into the cache. Amazing! Footnotes next =>
Thread (2/5)
Mat-Mat-Mult is O(n^3) if done naively on O(n^2) data that already hints there is some potential for data re-use. Now the funny thing is that even if the matrices are too big for the cache, you can still re-order the computation to make it fit into the cache.
Replying to @mtantawy
Also, if you really took this seriously, you‘d proxy the other website and make sure that it really loads slowly, not just the URL redirect.
Replying to @rorcde
I don‘t think the „this needs to be rewritten“ is based on any hard facts. I read that he had the same thing at paypal, and wanted to rewrite everything on Windows. Looks like the same kind of move to me.
Replying to @rorcde
Not sure, I heard they actually improved the infra and code a lot and were just getting ready to become more productive. Also, can‘t do a find & replace on strings in the codebase? Sure things can get difficult but this…
1
Thread (3/3)
@rorcde Still, if your data is already in there and you want to add some a bit more advanced retrieval capability, it definitely beats setting up a whole additional system and migrating and syncing your data!
1
Thread (2/3)
@rorcde - full text search is there, but using something like elasticsearch gives you a lot more configurability. - same probably holds for vector search.
1
Replying to @rorcde
After spending some time with it, it really is pretty versatile, and also has capabilities for full text search, and now apparently vector search. But I also experienced some short-comings: - for batch processing, it's still better to go with Spark/DuckDB/roll your own
1 View thread →
Still a lot of "tweet" lingo on this website (like the actual page). I imagine devs only change things when Elon comes by, points at the screen and says "it still says tweet here, change it." And then they change only that exact one location.
1 5
Replying to @_deep_yearning
I sorta could get 5 if I open the laptop... sometimes... when time demands it.
1
RT @guohao_li: This is promising for solving one of the most annoying problems in LLM - getting structured outputs. As one of the earliest…
Replying to @hkanji
Interesting, I tested myself a while back and it turns out it‘s just my tweets that suck :)
RT @FlaxSearch: Please RT! We need those papers to make #HaystackConf an awesome event - have you got a #search story to tell?
OK, people, be honest, can you immediately tell I'm a Gen-Xer because I use ascii smileys? (-_-') ;)
4
Alright looks like I need to figure out why the Internet isn‘t working so I can watch #jpnswe
1
Replying to @balazskegl
In German there is this idea of Sommerloch, literally a hole in the summer when not a lot is happening newswise and that stories that usually wouldn‘t become so popular get more coverage.
1
RT @shanselman: Beware the Prosperity Gospel of Tech Twitter. This whole system is algorithmically designed to make you feel bad about yo…
RT @year_progress: It's been a ride! There are currently 682,014 of you following this account, and we have been sharing memes, 69 jokes, a…
Replying to @rorcde
Hm, yeah, true those were released, too, but as you say they were already in the works. And the de-banning of accounts is more of but than a feature?! Depends on who you ask, I guess.
RT @_ahmed_othman: I couldn’t take this black X icon on the home screen, so made my own shortcut, welcome back my old friend https://t.co/…
Ok, so what did we get for 2.0? - view counters - verified got replaced with Blue - a new icon and edits on the string table Low key yay from me
2
Replying to @mtantawy
Yeah when you‘re this old, you have to take device churn into account 😂
1
Replying to @mtantawy
Did they ever update that number? Feels like I‘ve been seeing the same number since 1996!
Replying to @alung
Yeah, still works on web. ☺️ Somehow it seems all the work is focussing on the apps right now. Although I guess web would be much easier and faster to iterate on. But yeah.
I can't get how they replaced Tweet with Post in the app first while it is still Tweet on the website.
Man, I keep searching my home screen for this blue icon with the white bird and not finding it.
1 4
Replying to @tunguz
Just pulled the update, the Tweet is gone. At least for once they could have the balls to admit what atrocities they are doing in the update notes.
Media
Replying to @ridtalkstech
In German, one is „sciences of nature“ and one is „sciences of mind.“ I always found that „humanities“ sounded funny.
RT @sonicgott: You will forever be Twitter to me. I love this little bird. https://t.co/Wz9tnFh9sC
Media
Thread (2/2)
@paul_rietschka Future state of finance I guess 😅. Now that I just wrote this, I realize that the CEO of X didn‘t even mention the blockchain in her rebrand statement - what‘s up with this?!?
Replying to @paul_rietschka
This! 💯 If only crypto and web3 already sunk to the bottom of the ocean. Right now it looks like they're in a leaky raft still somehow floating.
RT @fmueller_bln: It’s gonna be amazing 🤗🤓 Together with your internal knowledge base in SO4Teams it’s a great pair programming companion.…
Replying to @krishnanrohit
@Suhail With people like Altmann I‘m really questioning just how much people reading capabilities humanity on average really has.
1
I can‘t shake the feeling Twitter is essentially group therapy for us all to unlock our inner sh*tposter and from my timeline, it‘s going pretty well!
7
Replying to @paul_rietschka
Yeah, ok, maybe I need to stop sh*tposting and read up on AmbiCond.
1
Thread (2/2)
@paul_rietschka Yeah HyperFlow sounds a bit 2016, and I think we‘re done with anything „EZ“ but AmbiSomething, why not. Given with all this talk about global warming it also feels like it was about time to make superconductivity at higher temperatures work. Gotta adapt, man!
1
Replying to @paul_rietschka
Ah, Keras! You meant other thread! Yeah no worries about TF‘s marketing here! ChatGPT suggests either AmbiCon, EzConduct, or HyperFlow as names for superconductivity at room temperature
1 View thread →
Replying to @paul_rietschka
There is a mascot? No, I was more thinking about „superconductivity at room temperature.“ We need something grippier like HoverTech or Superkool
Replying to @paul_rietschka
Also my impression. Which is mostly informed by people on Twitter, though. And the branding is terrible!
Replying to @paul_rietschka
Apple Silicon support! Love it! 😍 Finally a workload that gets the fans on my MBP spinning! Joking aside, I know it‘s not good for serious stuff, but it isn‘t bad either. Plus I think the direct memory architecture means you‘ll have more RAM than on a comparable NVIDIA GPU.
1
Replying to @paul_rietschka
@twiecki PyCharm tbh, but I love they are working on collaboration features!
1
Replying to @rorcde
@X Yeah, it‘s a good reminder In this particular case the user also wasn‘t superattached to the handle. But social media is a lot about compounding interests, building a following, establishing a brand (at least for many). The idea that this can all be taken on a whim is painful.
Replying to @smnbss
@X I'm probably concerned too much about my safety for SpaceX, but one of those flame throwers would've also been nice!
1
Replying to @markusandrezak
😳 holy crap, full on „nice brand you‘re having there. Would be really bad if something happened to it, wouldn’t it?“
1
RT @paul_rietschka: That was a fast update. With HF and **squeals like an excited child** multibackend Keras I’m sort of spoiled at present…
Replying to @gif_not_jif
Whoa, I don‘t think we‘re ready for actual feature innovation.
1
Replying to @tunguz
I still can‘t get over learning just being parameter optimization. Like all possible forms of knowledge are already representable if only you knew the right weights. Technically that‘s probably true even for humans, but conceptually, it seems lacking to me.
Replying to @WangUWS
@TeamBlind Interesting that wife trends so much weaker than girlfriend. Not yet ready to commit outside of your species?
Replying to @leonpalafox
Actually, I think what is really wrong about this is that least they should‘ve picked a proper accuracy metric. Number of weights alone says NOTHING (Unlike nuclear bombs, where it correlates with energy released).
RT @mikiobraun: @IgorBrigadir 99% of all researchers are doing superconductivity at room temperature wrong. But once you know the trick, i…
Replying to @IgorBrigadir
99% of all researchers are doing superconductivity at room temperature wrong. But once you know the trick, it is easy. Let me explain -> 🧵
1 1
Replying to @paul_rietschka
@untitled01ipynb That sounds great. Yeah landscape has its way too impress the mind…
Replying to @paul_rietschka
@untitled01ipynb Whoa, 3000ft elevation gain?! You did this as a hike or a run? 😅
Replying to @krishnanrohit
@shalcker But they have really great benchmarks for metrics like „ops per watt“!!
1
Replying to @paul_rietschka
@WangUWS Yeah, so many things they did first but then others copied. Face filters, stories, etc.
@holadiho Ungezählte Euros in der Rebranding-Campagne verschwendet, und es weiss immer noch keiner, dass die nicht mehr zu ebay gehören… das soll Dir eine Warnung sein, Elon!!
Replying to @paul_rietschka
@ajordannafa @Kyle_Ireton @realHollanders I did. I‘m looking forward to the first time he pulls and rebases and then has to fix all kinds of pseudo merge conflicts 😈
3
Replying to @oldJavaGuy
@alephnull42 Yeah it‘s the kind of joke where they have to explain it and then you‘re like well, yeah, ok, I see how someone COULD think this is funny.
1
Replying to @allenholub
Now Zuckerberg wouldn’t exploit this just to annoy Musk, would he :)
2
Replying to @alephnull42
🤯 It is just jokes all the way, right? This is all just a game to him!!
2
RT @alephnull42: I genuinely thought the unicode number was a joke - 1DG4F = IDGAF in 1337-speak. It seems to real - #Xitter https://t.co/W…
Replying to @CFDevelop
Yeah PRs are already stressful as they are, gotta stay as respectful as possible
1
Thread (2/2)
@paul_rietschka @untitled01ipynb Ok, new theory, Elon is the one running Linda‘s Twitter account, and he used chatGPT because he thought that would be funny.
2
Replying to @CFDevelop
@rickasaurus Exactly what I was thinking sitting here in my armchair nodding to myself.
1
Replying to @paul_rietschka
Uuuh googling it as we speak. Also, installing git on Windows was surprisingly painful…
1
Replying to @paul_rietschka
@krishnanrohit Anyone ever tested her profile pic whether it is AI generated? 😅
2
In other news I‘m helping my son implement a little game in Python to find connections with trains, and I can‘t believe I just reimplemented Floyd-Warshall for him.
6 View thread →
RT @MorningBrew: We need to stop all these rebrands until we figure out what's going on https://t.co/BC3L7i9OGl
Media Media Media Media
Replying to @paul_rietschka
Hey, I‘m also not in the target demo for NFT scams. Well, at least not anymore! 😅
Replying to @mtantawy
There was a time I was concerned about my online persona on here and wanted to appear professional, but we‘re just all the same now watching the spectacle.
1
Replying to @mtantawy
I honestly can‘t remember when the last time was I was talking about ML/Data on here in earnest. To everyone who is still following me 🙏🙏🙏
1
I have to say it‘s entertaining how a third of my timeline is scathing sarcasm about the rebranding, a third is people pushing their data/AI content as usual and a third is people burning through their startup funding running ads for crypto stuff.
11
Replying to @rorcde
Another of my favorites is restaurants telling you the card reader is broken today… 🙄 are fees really that high? Or is it all for tax evasion? I don’t know…
1
Replying to @rorcde
It‘s also that the shops are really pretty reluctant. Almost always there‘s a minimum amount before you can pay by card. And we have that EC card vs credit card thingy, too.
1
Replying to @paul_rietschka
It‘s a bit like when VCs say they invest in the team. This myth that some people can make stuff happen by sure willpower/elbow grease. I mean it is true that making stuff happen is a talent, but if it needs to work you also need to be in solid contact with reality.
1
Replying to @paul_rietschka
Yeah there is some exploitable gap of understanding between people with money and people who claim to have a vision where apparently you can get away with a lot of BS.
1
Replying to @paul_rietschka
Yeah, I still remember those photos from the launch pad. Completely destroyed, it just got burned to the ground. Amateur hour! The real problem is that people gave him a lot of money so he can just run his cos for much longer than would be advisable.
1
Replying to @rorcde
Oh it definitely was an overnight decision. I wouldn’t be surprised if Twitter‘s employees found out on Twitter.
1
Replying to @rorcde
Yeah, works for me, now, too. Probably DNS. I mean a real company would have redirected a couple of days ago so they already have control. But yeah.
Replying to @paul_rietschka
Yeah and then there is SpaceX… I mean I haven’t seen the numbers but is there an org with a worse history of launch failures?
1
Replying to @paul_rietschka
Yeah, none of that. It‘s also not really clear to me what happened. They seemed to be quite productive. I mean I know the stories about the punitive company culture but it seemed to work for a while.
1
Replying to @paul_rietschka
Yeah, it used to be a forerunner but now all the established car companies have caught up.
1
@BenHanowell Yeah, it has already started. Earlier today, X.com still redirected to a GoDaddy parking page, but DNS seem to have done it‘s thing now.
1
Replying to @_deep_yearning
As long as it is „hardcore.“ At this rate anything can happen.
1
Replying to @DRMacIver
Yeah it‘s less the bland visual nothingness that‘s this logo but that it‘s a reminder that we‘re just witness to a guy stubbornly plowing ahead with no regard to his userbase.
3
RT @hxiao: 🎉 Unveiling 𝗝𝗶𝗻𝗮 𝗘𝗺𝗯𝗲𝗱𝗱𝗶𝗻𝗴𝘀! A new set of high-performance sentence embedding models, boasting between 35m to 6b parameters, exp…
Thread (2/3)
@krishnanrohit No, I am 💯 with you. She switches categories ever two words, is it about interaction? Communication? Marketplace for ideas (whatever that is)? Services? And AI is just there because it has to be in 2023, admit it!
1
OMG, you can‘t make this stuff up, the redirect on X.com isn‘t even working. Amateur hour! At least put an IFRAME on there ffs 🙈
Media
1 2
Replying to @paul_rietschka
Yeah, naturally I’m not using it that much because Germany’s difficult relationship with electronic payment, but it is amazing how seamless it is. And I brandwise, we have one of the most high quality tech brands vs the shell of a tech startup.
1
RT @martingrasser: Today we say goodbye to this great blue bird This logo was designed in 2012 by a team of three. @toddwaterbury, @angyc…
Thread (4/4)
I agree with the article, the world has changed, online payment already exists, the premise that (maybe) made superapps a great idea in the past has changed.
Thread (3/4)
Of course there are examples of people who had to fight a lot of disbelief and eventually turned out to be right. But they still saw and looked for the signs. A superapp is just an idea that would be cool. I think in the cases where it worked there were reasons why it was needed.
Thread (2/4)
Smart people believing they have some secret insights and stubbornly ignoring all feedback is such a trope by now. But there is a difference between not listening to the naysayers and completely ignoring all evidence.
1
Replying to @kjhealy
@Ted_Underwood Honestly the amount of coercion that must go into her posting this nonsense makes me very uncomfortable.
2
Replying to @jstanier
I see some potential for a crossover with pirate stuff, like in „X marks the spot“ or „find X, find your treasure“ or something.
RT @ChombaBupe: Meta, Google, Microsoft & OpenAI have agreed to add a new watermarking system informing users when content is AI-generated.…
Replying to @francoisfleuret
I'd be OK with Feeling if it would be modelled properly. Why not. But it's not done. I think current LLMs maybe have some level of feeling as a latent state (or which could be read into what they are saying, but I think it should be modelled more explicitly, if we want it)
1
Replying to @allenholub
I've had to do ESTA a couple of times, but it was OK, very quick approval, ESTA is valid for two years (I think), etc. You can even pay via paypal :) Depends on how they implement it, it could be much worse of course.
2
Oh no, RIP Tony Bennet. His duo recording with Bill Evans was one of the first jazz records I owned.
4
Thread (2/2)
@rorcde And regarding Threads, yeah, the launch was pretty amazing but we have to see. I think pictures and video is quite a different medium from text, so it is unclear if the algorithms are transferrable.
Replying to @rorcde
Early to say maybe, as it indeed is a pretty big task. But I think time is also running out, and no major trajectory change so far. If at all, people are leaving slowly as they get fed up with all the drama about Melon.
View thread →
Replying to @garrett_wollman
@rorcde Ah yes, thanks for the reminder, I forgot about bundler!
Replying to @rorcde
@paul_rietschka I think „even Zuck doesn’t use it“ is really no indicator at all.
Replying to @mtantawy
Yeah my phone (iPhone 11) is still totally fine performancewise, too.
1
Replying to @mtantawy
Yeah I know Germans like to complain, but public transport in Berlin is actually pretty good!
2
RT @mtantawy: a moment of appreciation to the public transportation network of Berlin ❤️
Replying to @paul_rietschka
The amazing thing is that „none of this was predictable.“ I mean debt payments! Who saw that coming!?!
Replying to @mtantawy
Yeah for the last two iPhones I did the replacement after 2-3 years. Costs 79€ at the Apple store, still totally worth it!
1
RT @Noahpinion: Tokyo is to the early 21st century what Paris was to the early 20th -- not the world's financial hub, but its artistic and…
Replying to @alung
@HiCommunities I don‘t want to quote the Man, but this is indeed interesting!
Thread (2/2)
@paul_rietschka The absurd thing is that it could’ve been an aggressive move from end of June when there was all this noise about reddit. I have to admit my first thought was „whoa, this is a somewhat late but impressively fast move to pick up people who want to leave reddit.“ But no… 😅
1
Replying to @paul_rietschka
It appears this was released two years ago but the group was fired and now it seems to live in the unconscious underbelly of Twitter 2.0.
2 View thread →
Replying to @alung
@HiCommunities I see. Yeah that certainly didn‘t help… I don‘t know why it appeared now, definitely wasn‘t part of any communities yet. Maybe my A/B test group flipped.
Replying to @alung
Amazing, I thought I was following this space. (Pun not intended!) Either I completely missed it or they did a horrible job at PR.
Replying to @rorcde
Oh yeah, the pickaxe book! And that fever dream that was _why‘s Ruby comic guidebook! I wonder how they managed gem dependencies across projects like Python virtual envs. Did they?
Replying to @rorcde
Yeah, he is quite the number :) I met him personally a few times. He is actually pretty approachable and chill. Otherwise he‘d be unbearable 😂
1
Replying to @rorcde
Do you know his homepage? If you have the time, it‘s totally worth an afternoon deep dive. people.idsia.ch/~juergen/
Juergen Schmidhuber's home page - Universal Artificial Intelligence - AI - Deep Learning - Recurrent Neural Networks - Computer Vision - Object Detection - Image segmentation - GANs - Transformers with linearized self-attention - Goedel Machine - Theory of everything - Algorithmic theory of everything - Computable universe - Zuse's thesis - Universal learning algorithms - Universal search - Kolmogorov Complexity - Algorithmic information - Super Omega - Speed Prior - Independent component analysis - ICA - Financial forecasting - Evolution - Reinforcement learning - POMDPs - Reinforcement learning economy - Hierarchical learning - Metalearning - Learning to learn - Self-Improvement - Genetic programming - Attentive vision - Active exploration - Theory of beauty - Theory of creativity - Theory of Humor - Facial Attractiveness - Low-complexity Art - Lego Art
people.idsia.ch
View thread →
Replying to @rorcde
@rasbt @headius I mean most python projects also download a lot of packages, but most JVM build tools are just not lightweight (enough).
Replying to @rorcde
@rasbt Yeah, and jython got stuck being more of a prototype. @headius put (and is putting) years of work into making JRuby great. And I also think the JVM ecosystem is still too heavyweight for most data people. Now the only thing I miss in Python is proper multithreading. :)
Replying to @rasbt
Yeah, that post surprised me. I think it's perfect as it is. You get the best of both words: Near native performance but the flexibilities of a scripting language.
View thread →
RT @francoisfleuret: Me when the grad students ask if our new deep-learning thing is gonna work. https://t.co/kQ5F5sPynZ
Thread (2/2)
@francoisfleuret But with data, a lot is unknown and you won't know before you started working with the data. You know what has worked with similar settings before, but still so much can be different, data quality, distribution, seasonality, etc., that there is much more uncertainty than usual.
1
Replying to @francoisfleuret
First, great movie, second, yeah, I think that's probably the most important mindset shift people need to internalize when working with data. People often focus on algorithms and think it's a simple application - algorithm kind of relationship.
1 View thread →
Replying to @mtantawy
Yep, that‘s what we‘re about to go through for the next few years.
1
My various domain purchases were the other thing I was confronted with collecting my tax stuff. And then taking a hard look at myself and asking whether this was business related or not.
2
Ha! Collecting my tax stuff I saw that about this time 2 years ago my trello board, that I'm never using, renewed. Finally cancelled it!
2
Replying to @tunguz
Actually, in China it is one of the few unblocked western sites (they cannot, unless they want to re-invent all that open source software), so people are in fact using it as a social network.
19
RT @fedistats: Just released the new & improved trend detection. It's now based on how much a hashtag has gone up or down in rank from the…
Replying to @krishnanrohit
Yeah, maybe he was right that running Twitter required much less people, but creating new features still does. Currently the idea seems to be that "this isn't just properly monetized", hence the API costs and trying to get advertisers back on the platform.
1 View thread →
Thread (2/2)
@krishnanrohit That was a great thread! I'd say what's really been missing is any new features to speak of. I mean really, what's new? View counts, Twitter blue, anything else?
1
Replying to @leonpalafox
It was! I had them recorded some of our workshops. Those were the days! Seeing that lineup brings back memories! :)
1
Replying to @catalinmpit
@alisohani Nobody is talking about the typo in the headline? I honestly thought this is a different way to say „love language for seniors“ for the first minute!
1
Replying to @lemire
Three engineer-years, so could also be 36 engineers for a month! Or 1095 for a day!
5
Replying to @rasbt
@llanga @Meta Mark said we should do anything to make Threads work, so yeah… 🤷🏻‍♂️
Replying to @rorcde
Okay, if an LLM can sort out Scala version conflicts, I‘d be really impressed! :)
Replying to @fmueller_bln
@owehrens Internet midlife crisis! Finally I have a word for what I‘m in (just joking, it‘s a full blown midlife crisis!) But yes! 💯 I feel you! Actually I want something like open social media sites where you can bring your friends and the main act are RSS feeds. Over XMPP! 😍
RT @fmueller_bln: I miss the 2000-2014 internet time when we had blogs, RSS feeds, and Twitter had fail whales.
Replying to @alung
Waa, replied to the wrong tweet… anyway, alternatively it could be some open web standard for social media friends or something.
Thread (2/2)
@alung Yeah, I don‘t think GDPR is the reason Meta is reusing the follower graph 😅 But yeah, I mean the reason everyone is asking for the address book is that that‘s a data entity that exists on each phone - for obscure historical reasons.
Replying to @alung
Ok, so I thought about this for 10 minutes, and I think it should be asking for my „connections book“ because my address book is too personal.
View thread →
Replying to @AdamSinger
I think all of the young generation disagrees with phones being less ergonomic than „workstations.“
Replying to @alung
Hm, okay, also true :) But then they do this weird thing when you sign up with them, they already know who your friends are!!!
Replying to @ravi_mohan
Character limitation has always been the „appeal“ of Twitter. And the reason arguments get out of hand because everything is unnecessarily un-nuanced. Yeah.
1
Replying to @rorcde
@predict_addict To be honest, I wasn‘t aware Hastie/Tibshirani had any code in it. That must‘ve been introduced at a later edition!
Thread (3/4)
This whole bootstrapping your connections? Boooring! Figuring out who to follow to have sole background chatter? Exhausting! Someone with VC money for marketing build this, please!
1
Thread (2/4)
So I think they got federation wrong. It‘s not about reading the same stuff everywhere, it is about taking your connections somewhere else. It should be like you bump into people and say „fancy I‘d meet you here.“
1
Ok, hear me out! I think social media sites should be like bars/venues/stadiums/museums. You go there with friends/colleagues/Internet acquaintances to complain about the furniture and comment on stuff that‘s going on.
3 View thread →
Replying to @djpardis
The instagram mobile website is a pretty faithful representation of the app, so technically it could be done! But it‘s much harder to gather all that personal data, so maybe not?!
1
Replying to @mattturck
But it‘s also that on LinkedIn, followers can see your likes (I think), so they get a lot more exposure.
1
Replying to @ibogost
I would walk around reading a book. Looking down at rectangular things seems to just be in my nature :)
2
Replying to @rorcde
Yeah, no idea why Meta put resources on this, to be honest. Maybe still that feud between Zuck and MElon?
Replying to @paul_rietschka
@fedistats What was Millenium? It seems it's an amazon audible exclusive... . Meh.
Replying to @paul_rietschka
@fedistats Oho, someone turned this into an audiobook! Sounds very interesting!
Hypothetically I would jump on any new social media app (except that right wing crap), but luckily, the EU has put up enough privacy barbwire that Zuck would rather not. OTOH, Meta has several products out here and are probably very familiar with the requirements.
2 View thread →
Replying to @paul_rietschka
@fedistats Aha, another sci-fi author I should put on my reading list! I do honestly believe that sci-fi has something to say about exploring our relationships to AIs!
1 View thread →
Replying to @paul_rietschka
@fedistats Apparently it is an excerpt from an upcoming book called „Soul on AI“ by David Brin.
Replying to @sethrosen
A social network site with the login experience of microsoft/live/xbox.com, that would’ve been something!!
Replying to @krishnanrohit
Is superintelligence already more than AGI or on a completely different scale?
2
Replying to @krishnanrohit
Sorry, what’s SUPERalignment?! It feels like when they release their stuff they already went through several internal iterations of hype cycles.
2
RT @fedistats: I just released a complete overhaul of the fedistats website. 🎉 It's focussing on four areas now: - trends - search (new!)…
Replying to @ATRightMovies
I don't know how many times I've seen this movie as a teenager. Marty McFly was the coolest human being I could imagine.
1
Replying to @MCKettemann
@rorcde @anncathrin87 @BenFajzullin @dwnews Verstehe, vielleicht blieb das Bild haften, aber das Wort entfleuchte..
Replying to @rorcde
@MCKettemann @anncathrin87 @BenFajzullin @dwnews Trawling ist glaube ich eine Technik in der Fischerei. In dem Kontext macht es eigentlich keinen Sinn.
1
Replying to @MCKettemann
@anncathrin87 @BenFajzullin @dwnews Trawling? Die meinen Crawling, oder?
Replying to @rorcde
Yeah looks like a formula for keeping the users on edge emerges: hype topic + controversial figure + enough people you know to keep talking about both :)
1
Rate limit your Twitter. Rate limit your coffee. Rate limit your keystrokes. Rate limit your emails. Rate limit your credit card. Rate limit your alcohol. Rate limit your worries. Rate limit your carbs. Rate limit your road rage. Rate limit your air travel. Rate limit your twee
I love how the timeline is now half LLM/AI stuff and half the insanity that is Twitter 2.0.
3
Replying to @francoisfleuret
When you realize „this ain‘t rocket science“ means something different than you thought.
1 3
Replying to @rorcde
Yeah, I agree that‘s a very application specific skill set that merits it’s own specialization.
Replying to @rorcde
Ah, yes, I see what you mean. I have my doubts about prompt engineering specifically, but the way we‘re coding is definitely changing, and very quickly!
Replying to @rorcde
I do hope so! But the ecosystem is such a strong factor… it is incredibly unlikely to be successful at making a lot of people adopt a new (or old) language…
RT @tunguz: A lot of people are being hit by the Twitter rate limits, but 99% of them are not using them to their full potential. Here is w…
Replying to @rorcde
Yeah, I know exactly what you mean. The first version of TWIMPACT was essentially a rails app. Ruby felt like I could just write down pseudo code and then bit by bit refactor it into working code. JRuby was (still is!) pretty awesome.
View thread →
Replying to @wrede
Linda Yaccarino being "CEO" has strong "you're the CTO of a startup but actually you're just an inhouse coding agency" vibes.
Replying to @wrede
Then again, I just remembered he said he'll continue to "run engineering". Well, I guess it's time he needs to report to the CEO and explain to her WTF is going on. 😜
1
Replying to @rorcde
Yeah, I'd love to see Ruby become a viable alternative. There used to be Ruby wrapper for Spark. I think it's the cooler language for building DSLs. But Python's hold on the ecosystem was just too strong.
View thread →
Replying to @francoisfleuret
@VcWabKDHN9OVdYq No, he's mostly turning servers off or fires people who would write code.
What does the CEO of Twitter say to this? How does this help getting advertisers on this site?
3
Replying to @krishnanrohit
I'm still scrolling to get to the rate limits. First social media site that tells you to cool it.
1
Replying to @alung
You mean people go back to scraping now that the cheap APIs are gone? Yeah, that‘s definitely an unintended consequence!
Replying to @alung
Ok, so maybe it is really about the fears of being scraped and fueling the AIs? Reddit has been pushing the same narrative as well…
Replying to @alung
Aha, interesting. Well no friggin idea what‘s going on in his head 🤷🏻‍♂️
Replying to @alung
Yep, 100% my thoughts as well. It‘s infuriating how little he seems to know about how social media works. Have you seen the rate limiting? Unverified users can only see 600 tweets/d. Some say it is because he wants to reduce cloud costs.
View thread →
Replying to @alung
They turned off access for non logged in users. Not sure if they are allowing search bots but my guess is that they don‘t.
2
Replying to @CFDevelop
Yesyes, can‘t let that people get too engaged with your product!
3
Replying to @rorcde
@tunguz @MasterScrat I see. Still, I think if you‘d pour $$$ into it you could compete, I think.
Replying to @rorcde
@tunguz @MasterScrat Can you say why CUDA is so great? Isn‘t it all just linear algebra on the GPU?! 😅 Apparently I’m ignorant to the complexity of it all.
Replying to @mtantawy
Technically you could do this in Python, too, but for some reason, Python always wants to be more like Java and Ruby wants to be more like Perl.
Thread (2/3)
And before you say "uuh, maybe people will make an account to see the content!" The proper way how to do this IMHO you can see on instagram and tiktok where they show you a bit of content but then when you scroll down a dialog comes up telling you to make an account.
2
To spell this out: You find something you want to share, you share the link with your friends, they have a good laugh and think "maybe Twitter's not so bad after all, let me make an account" vs. this ->
Media
3 View thread →
Replying to @NoContextBrits
Man, I'd really love to share this post with my friends, but alas, Elon Musk is blocking previews.
Thread (2/2)
This feels a bit like reddit complaining that 3rd party apps are misusing the API. Individual users all using the same API key is not the same as mass data scraping! Don't be surprised if you have a huge social media site and the Internet is hammering you with request!
Thread (2/2)
@rorcde @tunguz @MasterScrat But yeah, I don't really get it. This is such a big $$$ market, it seems like if you have the resources you should be able to get to it and implement all the required computational operators, and do it well. Maybe AMD is focussing more on gaming?
Replying to @rorcde
@tunguz @MasterScrat Yeah I don't really get it either. I think one thing is branding. People don't want to spend a lot of money on something else and then figure out it's performance isn't up to par. I also often see people saying that implementations are incomplete, etc.
View thread →
Thread (2/2)
@rorcde I think what they historically optimize for are creatives, like having dedicated hardware support for high quality video codecs. But I think that is also quite recent, during their Intel era, they didn‘t really have that beefy machines except for the iMac Pro and the Mac Pro.
Replying to @rorcde
Yeah, weirdly they talk about it. When they presented the latest Mac Pro, they explicitly boasted with the amount of memory and that you could train the „largest LLMs“ 🤷🏻‍♂️
View thread →
Thread (3/3)
@rorcde Now that I start thinking about it, I've also seen people complain that it is near impossible to host your own email delivery anymore. Well, I guess everything has pros and cons...
Thread (2/3)
@rorcde The argument with Gmail was that they had a more global view and so on... But yeah, it's still not perfect, but good enough, I think. I remember people talking about that "spam would destroy the Internet."
Replying to @rorcde
I remember my university turning on "grey listing". It meant that they would reject the first request from any given IP. Normal mail programs would retry, but spam not. This made all the difference (but really disrupted email delivery for a few weeks).
View thread →
Replying to @rorcde
But you also remember the olden days? We spent days sifting through spam ads for Vi4gra and pen1le enlargement pills!
Replying to @hgggrA
Obviously a lot can go wrong! But that's no reason not to try. It could be more personal like an email spam filter, not like robot policing.
1
Replying to @francoisfleuret
Yeah, tokenization definitely gets in the way with this stuff. And also things like counting. It's ironic that we managed to take technology that the embodiment of bean counting and built something that cannot even count till five.
1 View thread →
Replying to @paul_rietschka
French painters? More like the Renaissance and metaphorical stinking or more like home renovations and chemical fumes kind of stinking?
1
It's 2023, I haven't seen a spam email in years and even the man on the street is using chatGPT, and yet we have to manually follow/block/mute/report our way through social media.
7
Replying to @tunguz
@MasterScrat Who said the startups are increasingly just intermediaries to funnel VC money to NVIDIA?
Thread (2/2)
@rasbt And sometime academia has this tendency to forget about paths it has already taken, so yeah. But linformers etc. are probably not too long ago. Then again given the speed at which the field is moving... Thanks for you thoughts, in any case, Sebastian!
Replying to @rasbt
Yeah, got it. It seems that we've pushed the envelope quite far in terms of performance, but I think there's also a real need for models that are still OK but can be run locally or even on CPUs, so I'm expecting some progress in that direction.
View thread →
Replying to @rasbt
Ah, I was under the impression that in LoRA you keep the weight matrices but have a low-rank update for fine-tuning, but maybe I misunderstood. Yeah, OK, the matrices are not like really low-rank, but maybe low-rank "enough" to do some compression...
View thread →
Replying to @rasbt
No, I mean factorizing the attention matrices themselves and replacing them with low-rank approximations after training.
Replying to @rasbt
I think there were some examples of replacing the multi-head attention with more low-level versions (localized attention, or low-rank matrix factorizations). In all applications of ML I've ever seen, no matrices were ever full rank (OK, I'm exaggerating), so it seems plausible.
Replying to @rasbt
I remember when Transformers first came out, there was already a flurry of work on making them more efficient (e.g. Linformers). Do you know what happened to that? Did it not stand the test of time, or is it time to revisit?
Thread (2/2)
@paul_rietschka @Carnage4Life It's sad we went from an Internet of snarky articles and ridiculous startup names to a wasteland of AI generated non-content and clickbait headlines.
1
Replying to @paul_rietschka
@Carnage4Life Wow that article had quite the attitude :) Also, it took OnlyFans to show that paid content works. What does that say about humanity?
Media Media
Thread (2/2)
@paul_rietschka And yeah, as people say, no DMs, no support for video or gifs. And as the user base got bigger issues with moderation etc. I think in a way it is also too similar. The question is what is playing over there? I know of one big account (@Carnage4Life) who moved over, but who else
Replying to @paul_rietschka
Yeah it does feel a bit like they got tangled up in the detail. Then again I don‘t know what their original speed was. I think they were working on this for quite a while already. It looks like they are still doing stuff, but I think the only big user facing change was timelines.
View thread →
Replying to @paul_rietschka
Ah ok, yeah, they really leveled of. They were already at 50k when I joined. Which was like an eternity ago. I mean 2 months. It was just 2 months 😅
Replying to @paul_rietschka
Ah, has it leveled off? Or you mean because they‘re still in closed beta? But yeah, it looks like the hype died down a bit. I‘m not even afraid anymore to openly say I have 3 unused invite codes!
Replying to @paul_rietschka
Yeah I heard there‘s a new venue in town! This place here had it‘s ups and downs but the new owner and the crowd… not what it used to be!
1
Over the last few days, accounts with totally non spammy names like Richard3748176 added me to lists called „BTC drop“… fighting bots is going well… And to the bots: catch up with the times, at least hand out OpenAI API keys or something!
5
Replying to @krishnanrohit
Well at least he‘d be a much better choice at hand to hand combat.
1
Thread (2/2)
Wondering what Yann LeCun is doing there. We met a couple of times while in academia but I don‘t think he ever replied on Twitter. True story: Yann once said to me „you‘re the only person I know who is actively using Google+!“
2
Replying to @rorcde
I'm guessing something broke. But yeah, others have also been guessing that this is the first step towards deleting inactive accounts...
1
I'm seeing a lot of people complaining their account has been blocked for violating Twitter's spam policy, often from accounts who are inactive or post like once per day. What's going on?!
1
Replying to @clairikine
The Regenband looks a bit too puny... Let's see whether it can overcome the rocky hotspot in the middle of Brandenburg that Berlin is.... .
1 View thread →
Replying to @clairikine
This morning when it was just 23C it already felt like 30C. So now that it is 30C… 😓
1
Replying to @mmitchell_ai
@timnitGebru @IEthics @jackclarkSF @BritneyMuller @tsimonite I‘ve seen enough from him. Blocked.
1
Replying to @americanwombat
I‘ve played Left4Dead years back and never made the connection and how odd it now sounds that we would indeed yell „Boomer!“ to indicate one of those are around.
Replying to @martingoodson
@ylecun I once had a chat with Schmidhuber where he entertained that we‘re all just living in a simulation, so it‘s not that far off…
1
Replying to @paul_rietschka
Oh yeah. Also, I seriously doubt „prompt engineer“ will be a thing of any permanent value.
1
Replying to @paul_rietschka
Now we‘re getting to the bottom of the existential crisis that is data engineering!! X-D
1
Replying to @paul_rietschka
Yeah, maybe if you don‘t want your data in GCP, I dunno. Yeah I mean each tool is slightly different and might have an advantage in certain conditions, and then there is people preferring their favorite tools and so on, but yeah, the whole hype cycle doesn‘t add anything useful.
1
Replying to @paul_rietschka
Yeah, yeah. One for in memory, one for batch processing, one for random access. The amazing thing for me has always been how in the end it all boils down to relational algebra. They really solved this in the 70s.
Replying to @paul_rietschka
Wait a second… people who believe in it… future profits… it IS like crypto!!!
Replying to @paul_rietschka
Wow. I‘m… processing. 🤯 I mean it‘s part of the playbook for open source startups. Because obviously it‘s not about revenue, so it‘s about how many people are using it, the size of the community. In the hopes that can then be turned into hard cash in some distant future.
1
Replying to @paul_rietschka
Yeah, and yet I never saw the connection to crypto scams this forcefully laid out 😅
1
Replying to @paul_rietschka
Regarding duckdb, I find that I have internalized the workings of SELECT … COUNT(*) GROUP BY 1 so much that I prefer to roll my own processing frameworks off of orjson in pure Python.
RT @WDAI_festival: Join our engaging panel discussion with Dr. Marielle D., Oksana Rasskazova, Dania Meira, and Morena Bastiaansen! Togethe…
Replying to @paul_rietschka
Not pictures in particular but yeah I saw the posts. He is definitely going through some anime style physical level up transformation. Like he made a pact with some demon or something.
2
Sooo, my memory‘s a bit shady but didn‘t Elon already agreed to fight someone a couple of months back?! This site… I‘m telling you!!
3
RT @fedistats: Working on a new UI, tab bar is already there but not functional but already I find myself tapping the buttons although I kn…
Replying to @bocytko
@rorcde The article from the guardian has a note that there was a clarification from Springer that the 200 job cut is from a restructuring of the regional news department, not from AI.
Replying to @paul_rietschka
@SashaMTL @pcastr You mean @untitled01ipynb? I just learned that exists!! I‘m out of touch!
2
Replying to @paul_rietschka
@SashaMTL @pcastr Aw man. I got Andrej Karpathy to unfollow me, where does that stand?
1
Replying to @AJButton2
And Meta is not even selling AI services! Re: the leak, honestly not sure whether they allowed it or they were still focussed on the Metaverse and it went under the radar.
1
RT @rasbt: I often get requests to dispel some of the jargon behind transformers and LLMs! So here we go, my new article on "Understanding…
Replying to @paul_rietschka
Have you already heard the joke that in the Matrix movies, they didn't keep humanity alive for the energy, but for the data they produced while jacked into the matrix. Well, not anymore it seems!
Replying to @touisteur
@bernhardsson I see. Yeah, I‘m not that deep in, unfortunately, to grok the differences.
1
Replying to @GergelyOrosz
I once Google Mapped to that city in Virginia that is home to a dozen data centers or so. It‘s geographically incredibly packed. Like an Amazon fulfillment center.
Replying to @bernhardsson
First time I saw it. Intel GPUs? Those embedded GPUs? Also got me wondering why AMD has no drop-in CUDA libraries. IIRC they use some other API.
2 View thread →
RT @krishnanrohit: Large chunks of equity raises used to go to: - Fb and Google for ads - CRE for rent - AWS for infra spend - Now, GPUs ga…
Replying to @rorcde
Ok, according to crunchbase Immerok got $17M funding, one order of magnitude less. We’re definitely on a different hype cycle right now.
Replying to @rorcde
I find the reporting a bit odd. „Only four weeks old?“ But I guess they have been out raising this money for a while and only then incorporated? Also it seems like they gave away 50% or so?
Replying to @paul_rietschka
@HamelHusain @eigenhector Yeah, I've been using an MBP with a M1 Max and that thing is so overpowered. Makes my dual core Intel MBA feel like a raspberry pi... 😅 (I still love the old MBA form factor, though. I'm typing this on the MBA, don't feel bad, buddy, you're still beautiful!)
Replying to @paul_rietschka
@HamelHusain @eigenhector Yeah, and I also think that Apple Silicon was mostly targetted at mobile devices (and I think it's truly a game changer there). Given that I think it's pretty impressive how much they manage to scale the performance up.
1
Thread (2/2)
@paul_rietschka Maybe we can ask sama. Somebody told me he had hard numbers for the impact of Italy banning chatGPT on developer productivity. If anyone knows, it‘s probably him.
Replying to @_deep_yearning
@harsh6363 For a moment there I was worried they went out of business… 😅
Replying to @krishnanrohit
Spoiler alert, but that‘s literally how it started in Horizon: Zero Dawn! Can people please consume more scifi content before they try stuff like this?!
1
Thread (2/2)
@paul_rietschka @NeptyneHQ But yeah, I mean lets face it, one of humanity’s duties is feeding our AI overlords with data. It’s about time we start taking this seriously. We can‘t keep wasting CPU cycles on parsing malformed data like this!
Replying to @paul_rietschka
I totally agree and this is exactly the kind of product which would be very useful but doesn‘t differentiate enough from the status quo and probably never gets built :(
View thread →
Replying to @eigenhector
@HamelHusain That benchmark in particular puts the M2 against CPUs with more than double the number of cores. If you look at the numbers of the i9, it is much more on par. So it‘s not wrong, but the argument is more about there not being versions of the M2 with 50+ cores.
1
Thread (2/2)
@paul_rietschka I guess it /can/ be an ok tool for data entry that almost everybody can use. But it can definitely also be a horrible tool for data entry!
Replying to @paul_rietschka
Yeah I guess you *could* make it to be mostly tabular data, but in reality… So I‘ve seen pretty horrible things with Excel-in-the-pipeline and I‘ve seen people being quite enthusiastic about hooking up Google Sheets for data entry…
View thread →
Replying to @Quasilocal
Second to last place? My knowledge of the arcane art of academic signaling is a bit rusty, but isn‘t that the PostDoc who actually supervised the work? A steal!
4
Replying to @bocytko
Ah yes, that makes sense. Regarding the post mortem, let's see, but I doubt it...
RT @fedistats: Alright, I held off all weekend, but now it‘s time to deploy the big pipeline rewrite so that the notifications are sent out…
RT @tdinh_me: What a read. As platform risk is the topic these days, I highly recommend reading this. Learned a lot of things from this.…
Replying to @rorcde
Yeah, definitely, a lot of consolidation happening. Re: social media, facebook has been doing very well with ads (and other shenanigans 😅), but I'd also be fine with it being paid. As the old saying goes: If you aren't paying for the product, you're the product!
1
Thread (4/6)
The M1 Max GPU on my MBP benchmarks a bit slower than the GTX 1660S on my PC. That's not that impressive, but the new M2 Ultras have more than double the cores, so that would put them somewhere along the xx70 series? (assuming roughly performance doubles from series to series)
Thread (3/6)
Re: the latter, Apple has been working hard, for example, providing the mps backend for pytorch, but most people still stick to Nvidia GPUs for training.
In Apple's Monday presentation, they mentioned a couple of ML related things. From what I remember: - transformer models for speech2text and autocorrect running on the device - they claimed that the new M2 Ultra with 192GB unified memory could be used to train LLMs.
1 1 View thread →
Replying to @rorcde
Oh yeah, we totally live in a bubble subsidized by VC investor money, especially in tech. But also look at food delivery or ridesharing, I think these are also heavily subsidized. And that's not necessarily a bad thing. But yeah, at some point the money runs out.
Thread (2/3)
@rorcde Not totally alienating your ecosystem could also be an alternative. It seems these new prices were given with very little advance warning and discussions have not lead to anything.
Replying to @rorcde
If these are the prices they absolutely need to be profitable, than that's a problem. I also think there is a difference between 3rd party apps API and pure data access (although also a 3rd party API could also be used for data retrieval).
View thread →
Replying to @InkmiHq
@KingOfCoders Ha! Alles nicht so einfach! Man kann ja nicht einfach noch einen zweiten Rechner dazustellen! 🤦🏻‍♂️
1
Replying to @francoisfleuret
@cthorrez „Text based faces“ That‘s what it has come too!! 😅
RT @lilianweng: 🚜 Cannot believe it is almost 3 years since my 2020 post on variations of Transformer. I spent some time and did a big refa…
Replying to @ungatedlife
@ungatedcreative All the while I saw you can write you own custom timeline service over at Bluesky. I haven’t fully understood how that works, but it could be really really cool.
1
Replying to @francoisfleuret
But look at the power of the person that came up with the five letter acronym! True story, my professor would spend at least 30% of a project proposal‘s writing time on tweaking the title till it gave a good acronym.
1
I keep repeating that AI is dangerous when we hook it up to systems where it can do real damage and we let it. Autonomous cars getting in accidents. Data driven algorithms that make bad economic decisions. AIs in weapon systems. AI is not the problem but what we do with it.
2 10
Replying to @vivekjuneja
@Plinz Hey there, unfortunately I‘m busy, but maybe we can connect in other ways!
RT @ayodele_odubela: The fact that white men in AI are playing up the “aloof, regretful scientist” trope is ridiculous when they COMPLETELY…
Replying to @paul_rietschka
I've said this before and will repeat it till PEOPLE WILL LISTEN, but I think AI energy consumption as a main driver of climate change is one path how AI will actually bring about the extinction of humanity!!!
Replying to @paul_rietschka
Aren't some people now also saying that AI is what will help us solve all those problems so we should rather focus on AI than, say, climate change?
Thread (2/2)
@paul_rietschka Along those lines, I had to LOL when I saw that 22 word statement deciding that the biggest threats to humanity are AI, pandemics and nuclear weapons. Who made that short list?!
1
Replying to @paul_rietschka
Yeah it's really weird and definitely not a good picture. I wonder whether the media is also amplifying it. Or men are just drawn to doomsday scenarios and lose their common sense.
1 View thread →
Holy crap, one more year and my Twitter can buy a beer (in Germany) #MyTwitterAnniversary
Media
4
Replying to @ridtalkstech
I really don't know why we need to have a competition of extinction risks now. It looks like we're enough people to divide and have enough people working on each. Fair enough, I don't think most people would see it as an either/or, but Hinton does apparently.
1 View thread →
Replying to @hardmaru
@Meta @NVIDIAAI @NVIDIAAIDev @GoogleAI @amazon Not fully sure what ensconced means even after looking it up, but I don‘t see Apple or Amazon being hot in any way wrt AI.
Replying to @paul_rietschka
Three weeks ago? You mean in April?!? Man I remember that April like it was yesterday…
1
Replying to @ridtalkstech
Oh yeah, I think implanting anything in the body is much much more challenging than scifi makes you think. I'm also immediately reminded of an article I read about medtech startups that went out of business. Wouldn't it be great you'd be left with an EOL'd implant in your brain?
Well, almost got it right... . But then it seems like it just preferred to get an integer solution in step 5... .
Media
1
Replying to @markscott82
Not just digital policies. Some AI doomers are explicitly saying AI is more important than climate change.
2
Replying to @ridtalkstech
Last I heard rural German still has terrible mobile Internet 😋
Replying to @paul_rietschka
Alright so we reached that stage where the illuminous leader is but a myth and you regularly see people who say they saw him and he‘s in great shape!?
Replying to @paul_rietschka
Biden already looked old when he became president but in recent pictures it looks like he‘s aged another 20 years since. The other day Mark Warner was on the Pivot podcast and they emphasized he‘s one of the young ones, bringing in some fresh energy. He‘s 68.
1 View thread →
Replying to @paul_rietschka
Yeah… I can‘t shake the feeling he‘s just not in a good space mentally and all this shouldn’t be as public as it is. Then again people forget how eccentric scientists sometimes are…
1
Replying to @krishnanrohit
Yeah, reading the docs first I was like „wow, this is huge, if it works, but it feels like quite the stretch.“ Only relying on natural language descriptions and have the AI „figure it out?“ How do you make sure it works? How do you tweak it?
Replying to @srchvrs
Yeah, looks interesting if true. But the wording… why did they have to call it „imitating“?! That alone sounds like it‘s not the real deal.
1
It took me literally a minute to understand how the words „Geoffrey‘s MacBook Pro“ fit into the picture.
4
Thread (2/3)
Some first thoughts from me: - you don't need to understand the math - stick to Python - understand how ML is not yet another library - learn to properly evaluate models
How do you get started with AI as a Software Engineer? I know there's a lot of talent out there, but most material is way too heavy on the math. But what do you really need (and don't need) to know?
View thread →
Replying to @leonpalafox
That's the good kind. The other two options are scorching hot 37C out of nowhere or 20C and rain for days.
Replying to @_deep_yearning
Yep, I know… can be pretty useful but totally underadvertised.
1
chatGPTllucinating: when you make up stuff just because it sounds plausible but cannot remember any source or evidence.
On the positive side, looks like we're set up for a week of sunshine and temperatures in the low 20s (Celsius!) here in Berlin. Marvelous!
1
Replying to @paul_rietschka
I‘m so old school I‘d even spin up an SVM when I feel like it.
Replying to @paul_rietschka
Looks like I finally said something sufficiently controversial. 😅
1
Replying to @Henrikop
Yes, it depends on your use case. Some stuff can't be done without LLMs. And I'd turn your last sentence around and say "more energy is only justified with much better results!"
Thread (2/2)
Sure, you can do document classification with an LLM and it'll probably work for much broader documents, but running an LLM requires a machine with a lot of GPU compute power, while you can run a xgboost model on standard hardware, for the fraction of the time.
5
With everyone is focussing on LLMs, I'd like to remind people that "classical" methods like xgboost are as relevant as ever even for text based use cases (e.g. sentiment classification, document classification) because they use a fraction of the resources that an LLM does.
4 16 View thread →
Replying to @d_stepanovic
Yes! There's even a mathematical proof, because the Taylor approximation at x=0 is exp(x) = 1 + x + ...
2
RT @omarsar0: QLoRA - an efficient finetuning approach that reduces memory usage enough to finetune a 65B parameter model on a single 48GB…
Replying to @mhartington
@reactdayberlin @Ionicframework God, now thinking about it, I hope it's not some kind of facial paralysis... 😰
Alright, I finally figured out how to get more people to use @fedistats! Rewrite the data pipeline in duckdb!
1 2
Replying to @paul_rietschka
Yep. It seems when transformers first came out there was already a flurry of efficiency research. That plus open source models will make for interesting times. Hopefully.
Thread (2/2)
@paul_rietschka I stumbled upon that because LoRA kept talking how very low rank changes can make significant improvements and so on (not surprised about that tbh) So yeah, always seeing the best in people I suspect they kept increasing model sizes for marketing purposes.
Replying to @paul_rietschka
Yeah it‘s wholly impractical. I was reading papers (yeah… I know) and stumbled upon linformer… it uses low rank matrix approximation on the attention matrices… whatever happened to that?
View thread →
Thread (2/2)
@paul_rietschka Model sizes have become meaningless like car model numbers. In a couple years we won‘t even remember that the B stood for billions.
Replying to @gasnerpants
I still find myself trying to find that special kind of 90s house music, the one with the organ bass. In my mind, that was fun music. Can someone bring that back?
Replying to @mucio
@fedistats :D So I tried some things and even if I delete some data it doesn‘t get much faster. I think ultimately I should do the kind of aggregates I‘m doing on postgres now offline.
View thread →
RT @mmitchell_ai: I've mostly not spoken up about longertermism, effective altruism, and AI. But when it comes to affecting what we priorit…
Replying to @mucio
@fedistats Not much in terms of data engineering, just some Python scripts and a Postgres database 😬
Replying to @paul_rietschka
There are people in Germany who pronounce Jazz as „yats“ unironically so yeah.
1
Welp, the hourly job at @fedistats is taking longer and longer to run. Looks like I need to start deleting old data from the database 😅
Replying to @paul_rietschka
Great song! Didn‘t know it! Now with post rant clarity, I think what I really want to say is that there are no short cuts and whatever tool you use, you will need to learn how to use it well. LLMs won‘t change that. You get in at a different point, but you still need to learn.
1
Thread (3/3)
@paul_rietschka This is not just me who did put in the hours and now I can‘t have the young folks get something without the pain I went through, this is the way our own brains work I am afraid. End of Gen X rant.
1
Thread (2/3)
@paul_rietschka And it is true, you can crank something cool sounding quickly. But then it invariably turns out that to go beyond that you need to get back and learn the basics, put in the hours, and then you got what you need to use the new tools to do something great.
1
Replying to @paul_rietschka
Yeah, so let me be the Gen Xer I am, and we‘ve all seen this before, maybe on a different level, but if you take music, there were synthesizers, drum machines, MIDI packs, loop libraries, and every time people were like „wow, you don‘t need to learn an instrument anymore“
1 View thread →
Replying to @paul_rietschka
Yeah totally. Like that „synthesize your voice from 3s of audio.“ I saw people are now pitching it as a way to help disabled people, but this is one of the cases where I see overwhelming bad uses.
1
RT @pramsey342: Calling it “content” was the first step in convincing people that machines could make it
Replying to @rorcde
Yeah, I see what you mean. I also found myself being a bit taken aback till I realized it‘s my own newsletter 😆
1
Thread (2/3)
@rorcde For example, this week was a lot about AI regulation. That was new! I threw the headlines into chatGPT once to see how it would do with summarization, but it actually became even wordier :)
Replying to @rorcde
Very happy to hear that! 🙏 And yeah, there‘s definitely stuff that can be done on top. Luckily there is very little duplication right now, but that‘s just a coincidence. And yeah, some more info on topics would be interesting.
1 View thread →
Replying to @rorcde
Ah, a very disciplined approach to social media :) Yeah, for that case I can see it adds unnecessary complexity!
Replying to @rorcde
You check your email only twice per day? Have you turned off email notifications? And you mean the fedistats summary works for you? Happy to hear that (if I understood you correctly :))
Replying to @rorcde
Yes. You get to pick your time. Substack etc. don‘t have it because the creator picks the time for you. I‘d also prefer a simplified sign up, but an early feedback I got was that it‘d be nice to pick the time.
I’m amazed how thoroughly un-engineering-y prompt engineering is right now. It‘s just a bunch of stuff people have found to impact the types of answers, but neither do we know how nor are there any guarantees.
4
Replying to @stefan_will
True! Now let‘s see whether Zuckerberg is able to steer Meta towards something as mundane as cloud services.
1
Thread (2/2)
This is literally every engineer (including myself) creating his own private toolset because they believe that it will help them be so much more productive, instead of focussing what they are actually building.
2
Meta thinks it is on the same level as Microsoft and Google, but the truth is that Meta is a social media/ad company. It does not have a cloud business where it could easily turn this tech into money. techcrunch.com/2023/05/18/met…
Meta bets big on AI with custom chips -- and a supercomputer | TechCrunch
Meta wants the world -- and particularly investors -- to know it's going big or going home where it concerns AI and the hardware to build it.
techcrunch.com
3 View thread →
Thread (2/2)
Although if I understood correctly it introduces the idea of going from supervised learning to learning an unsupervised language model without which the whole prompt engineering business would not have been possible!
OMG, I'm reading papers again. What's interesting is that until a few years ago it very much sounds like scientific business as usual. Even the GPT-2 paper "Language Models are Unsupervised Multitask Learners" sounds relatively level headed.
1 3 View thread →
Replying to @paul_rietschka
God, no! Metric all the way, you should try it too! True story, every time I come to the US and see a digital clock saying something like 3:45 in the afternoon my brain goes "but wait, it's bright outside, it's not in the middle of the night!"
1
When you live in a country with 24 hour default and you try to be more like the US but still fail. 😂 I had this on for at least 3 weeks before I noticed it.
Media
3
Replying to @mtantawy
This is even worse than making the compiler happy. At least there were rules!!!
Replying to @antonioalegria
So maybe they are bound on compute resources… or they just take any trick from the playbook to make their stuff feel bigger and more complex than it really is.
Um, what‘s the excitement about the chatGPT app for iOS? I mean better integration is great, but are some people assuming chatGPT now runs on your phone?
2
Replying to @ParvSondhi
@deployonfridays @JustAnotherPM Friday mornings, pah, wait till 4pm!
2
A friend told me he had used chatGPT for brainstorming. At some point I had to stop him to ask him why he kept referring to chatGPT as "her." I don't even know what this says about humans an bias.
1
BTW I love the fact that I found a blue looking emoji (a jellyfish) that doesn't properly render in Safari.
Media
RT @goodside: Google Bard is a bit stubborn in its refusal to return clean JSON, but you can address this by threatening to take a human li…
I’m aware this is inaccurate because the congress never wanted to regulate guns in the first place.
Media
8
What is it called if it doesn‘t matter whether arguments are true because both sides already figured out what they want to do, and the arguments conveniently support that? That‘s the vibe I get from Sam Altman calling for AI regulation and the US congress going along with it.
1 View thread →
Replying to @hardmaru
What I‘m reading is that he calls for companies needing a license, thereby adding a lot of red tape and making it harder for the competition (especially open source?) to catch up.
1
Replying to @richardtomsett
@krishnanrohit Yes. I think it has a more concrete specific meaning in physics, but applied to something like behavior it is really unclear what it is (although I would also say I have some idea what is meant by it).
2
Thread (2/2)
@krishnanrohit IMHO the phrase „taught itself“ also gives the wrong image. FWIW, deep learning is more like learning a motor skill. Endless repetitions and somewhere along the way you get better.
1
Thread (2/2)
@rorcde Alright, you're set to go. The unsubscribe link in the email should work as well :) 🙏 So, you wanted to know the opening rate. If I'm reading this correctly, it fluctuates, but is something like 40-70%. ☺️
1
Replying to @rorcde
Thanks for your patience and sorry for the hassel :) The link doesn't work anymore, you have to register again, but let me just do this for you... You should get another email, but I'll activate it manually for you!
View thread →
Replying to @rorcde
OK, once an email is in, you cannot register again. I removed your unactivated signup and you can try again. I probably offer to send the email again :)
Thread (2/2)
@rorcde Ah, you had a "+" in your email address and I didn't properly urlencode it, so it became a space back. You want me to manually activate or you want to try again?
1
After pushing Geoff Hinton‘s anxiety attacks daily, I am happy to see that the Guardian also publishes more nuanced pieces like this with which I wholly agree. AI doomerism is marketing, AI anthropomorphisation is storytelling, and it’s all a distraction theguardian.com/commentisfree/…
The apocalypse isn’t coming. We must resist cynicism and fear about AI | Stephen Marche
Remember when WeWork would kill commercial real estate? Crypto would abolish banks? The metaverse would end meeting people in real life?
www.theguardian.com
3 11
Shit, notion's AI feature that summarizes everything you've written so far is really quite good and helpful 😅
1
Replying to @fmueller_bln
I wish I had tips but I‘m also struggling to stay sane 😅 That said, remember that the only thing that‘s real in life are people and at the end of the day, the only thing that counts. Until we have companies run by AIs, then you can just focus on making it happy.
RT @DrJimFan: AI Twitter is flooded with low-quality stuff recently. No, GPT is not “dethroned”. And thin wrapper apps are not “insane”. At…
Replying to @hardmaru
@marvinvonhagen Can‘t they at least passphrase protect these things? Like in old spy movies? The crane flew over the lake at night or something?!
RT @fedistats: The fedistats journey continues: Latest addition is an AI News Newsletter (I need to get better at wording these)... https…
Replying to @leonpalafox
@krishnanrohit @lawrennd Even packed with bodyguards, crazy! Well, given recent events looks like he leveled up and unlocked some hand-to-hand fighting skills :)
Replying to @leonpalafox
@paul_rietschka Yeah, so hopping threads, but I think that's exactly the problem with Meta and AI, it is too tangential to their business.
1
Replying to @leonpalafox
@krishnanrohit @lawrennd I heard he was flown in with a helicopter and had bodyguards in the room. Is that true?
1
Replying to @reactdayberlin
@Ionicframework I rarely engage with ads, but that's some strong eyebrow game, Mike!
1
Replying to @leonpalafox
@krishnanrohit @lawrennd So you were there! I was in the bitter phase of my PostDoc and felt it was too exhausting to travel to another continent for a conference. 😅
1
Replying to @leonpalafox
@krishnanrohit Oh yeah, that, too. Oh man, has it already been ten years?
2
RT @mikiobraun: @fmueller_bln Okay, here me out, so Google was the lone wizard who created it, OpenAI was the villain who stole it and want…
Replying to @fmueller_bln
Okay, here me out, so Google was the lone wizard who created it, OpenAI was the villain who stole it and wanted to have it all for themselves, and then Prince Mark was the one who freed it and gave it to the people.
1 2
Replying to @GergelyOrosz
OMG, Gergely, links!! The thing that almost destroyed the Internet! Everyone knows how dangerous they are!!!
Replying to @krishnanrohit
If my company went through what his company is going through I‘d also have a lot of pent up rage to tap into.
12
Replying to @rorcde
Not sure, I think that‘s also how he likes to be seen. Truth is he is director at a prestigious Swiss research institutes since the 90s and can do whatever he finds interesting. For some, this is already the dream.
Replying to @rorcde
It seems you haven’t been Schmidhubered. Joking aside he is notorious for sitting in talks and then asking people if they are aware a student of his already solved this „in 1997.“ He did some (very theoretical) work on universal algorithms, so technically he‘s not even wrong :)
1 View thread →
Replying to @paul_rietschka
Oh yeah, I meant really just claiming. I also see no way they could actually do something. They have good people, but as you say, they‘re a social media platform and an ad network, no cloud or SaaS or anything.
What I'm saying is that I see a real chance for Mark Zuckerberg to be claiming in a year that Meta were the ones who really started the AI revolution - but he needs to bury the Metaverse first, and quickly.
4
I think the big irony of LLaMa being what initiated the recent developments in open source LLMs and fine-tuning is that Meta probably would have never released it if they hadn't been so focussed on the Metaverse.
5
Thread (4/4)
I mean and also think about the fact that super focussed specialization is something that mostly benefits big tech companies. Everyone else needs people with a bit of range.
Thread (3/4)
And BTW, if you have test data and a metric, prompt engineering becomes yet another exercise in generative AI. Think about that before you go all in on prompt engineering.
1
Thread (2/4)
This includes "prompt engineering." This isn't like learning how to build an app in flutter. What they really should teach people is how to define problems in a quantitative way and how to properly validate on test data.
2
I think the way "AI" is marketed will lead to even more people wanting to use it who have no experience what it means working with data. Anything that is built from data is approximate. Don't believe it is solved because big tech tells you it is.
1 4 View thread →
Regarding the announcements for a new Chief Birdsite Executive Officer: I'll believe it when I see it.
I learned today that apparently chatGPT costs $700k per day to run. Talk about burn rate!
2
Replying to @paul_rietschka
Yeah, so much to unpack about tech in general. But I have to say I almost lost it when Nick Offerman said „it‘s big data“ 😂
1
Replying to @paul_rietschka
Oh yeah. And don’t let her talk you into stuff you then think were your own idea!
Just returning from the @M3_Konferenz and the amazing thing is that although we put the program together in January, we still managed to get enough talks in that could now pick up on everything that happened in AI and LLMs in the past weeks.
2
Replying to @GergelyOrosz
@holafly_com Oh yes, I have been using @airalocom for years (not getting paid to mention them), it used to be a bit hard to configure but got much simpler over the years and they have great data plans (no hotspot, no calls).
Replying to @paul_rietschka
Not there yet, Lily just crashed the car. Intense stuff!! I personally also enjoy the Asian American representation!
Replying to @paul_rietschka
@krishnanrohit Yeah, probably another couple of years till the funding is exhausted.
2
Replying to @fmueller_bln
Yes, definitely, hopefully the relationship is constructive and positive enough to allow for this. Some PMs think they tell the team what to do. Mentoring „up“ or „laterally“ might be challenging then.
Replying to @fmueller_bln
I hope you mean „as their boss“ and not „as an engineer in the team.“ 😅
„As an AI language model“ is their version of a software engineer‘s „it depends.“
2
Replying to @hardmaru
Yeah, obviously it has been told to say that, but what does it REALLY want to say?!
1
Replying to @krishnanrohit
It always amuses me to no end when they are asking for fiat currencies to keep their businesses running. I thought you said never trust fiat!
1
Replying to @paul_rietschka
Man, that‘s the part about „decentralized“ I like the most!
Replying to @paul_rietschka
Ouch. Well 20% might have been possible, you just need to know when to pull out. Like the exact second.
Thread (2/2)
@paul_rietschka Speaking of old men yelling, I‘m at the platform together with 500 other people waiting for the train which just got delayed by 10 minutes. Never a good sign.
1
Replying to @paul_rietschka
I still see people with that hexagonal profile pic. Friggin‘ losers. Who‘s past their prime now?!?
Thread (3/3)
@paul_rietschka The most amazing thing were all those crypto conferences that popped up, pulling in young engineers who thought they could become part of the next big thing. With hackathons and raffles and all that. Luckily covid put an damper on those events.
Thread (2/3)
@paul_rietschka Web3 was so amazing because it really lacked any substance. No actual tech looking for a use case, no customer problem to be solved. Just the promise of making a quick buck before people figure out what‘s up.
Thread (2/2)
@paul_rietschka Insane amounts of money. I was at some breakfast meetup on gen AI right now. Lots of talk about privacy, working business models, revenue, regulation. They outright said you don‘t see US style massive investments into technology in Europe.
Replying to @paul_rietschka
And also forgot all their startup roots and poured billions (?) into this.
Thread (2/2)
@paul_rietschka If that becomes more mainstream at some point I think that Microsoft with their AR are on a better path - although I never forgave them for not bringing out a VR headset for the Xbox.
Replying to @paul_rietschka
Yeah. I‘m also always concerned how many times that screen can be folded before it breaks. And I mean yeah, they couldn‘t predict that AI would take off now like it did, but then again the Metaverse was nothing but a nerd‘s fever dream from day one. Nobody was waiting for that.
View thread →
Replying to @paul_rietschka
Ehehe. And here I was asking myself whether that was actually Python. Focussing on the tech stack, as usual.
1
Replying to @paul_rietschka
Yeah, 💯 I have tried VR for gaming and it is cool but also quite exhausting and too immersive even. More than once did I try to lean on something that wasn‘t actually there. I think you can learn to be aware of what you see and what’s actually there, but why.
Replying to @paul_rietschka
Caught us by utter surprise. I‘m wondering, was it the legs?
1
Replying to @lemire
I think we would need to design these tests differently for GPT-4. It has much larger memory than humans, but should be challenged more towards transfer to entirely new settings, and be challenged with very plausible but wrong settings.
Replying to @wrede
And as you can see, I‘m not even re-reading my tweets before I send them 🤦🏻‍♂️
Replying to @wrede
I stop re-reading my (work) email years ago. You just type and then hit Meta+Enter. The only way to stay sane!
2
Replying to @fhuszar
It's like Anna Karenina but for the AI community: "All happy families are alike; each unhappy family is unhappy in its own way."
Replying to @paul_rietschka
Yeah, but what I really liked was that it was about crews and what is leaderships and what is the right thing to do etc. And yeah, no spoilers, but I think this was the last season with this cast.
Replying to @paul_rietschka
Oh, season 3 is awesome, much more an ensemble piece like the old TNG series.
Replying to @paul_rietschka
Ah, Allison Pill, thought she was great in Scott Pilgrim vs the World!
1
Replying to @paul_rietschka
Ah yes, I assumed you knew but then got carried away by how fitting the summary it is to the super rich about the dangers of deploying AI for personal gain 😅 And I have not and am delighted to see it is streaming in Germany on Disney+! And Nick Offerman!
1
Replying to @paul_rietschka
Yeah, although this is from Ex Machina, a movie where a super rich dude tries to create the perfect robot/girlfriend only that they all go crazy or *spoiler alert* kill him in the end and escape to the real world. So it‘s fitting in a way :)
View thread →
Man, the Guardian really is on a roll handing out meaningless titles. So Geoff Hinton is one of the three „godfathers of AI.“ and now meet the „father of AI“, Jürgen Schmidhuber theguardian.com/technology/202…
Rise of artificial intelligence is inevitable but should not be feared, ‘father of AI’ says
Jürgen Schmidhuber believes AI will progress to the point where it surpasses human intelligence and will pay no attention to people
www.theguardian.com
1 4 View thread →
RT @fedistats: You want one email with all the AI madness from reputable news sites in your inbox once per day at a time of your choice? 👇
Replying to @bobblakley
@matthew_d_green And put it on a Blockchain and post it to BlueSky and we have achieved MaxHype.
3
RT @fedistats: New feature! Now you can subscribe to a newsletter that is built from a fedistats topic. Subscribe to get a daily summary of…
Replying to @rorcde
Yeah social media is something different IMHO. I honestly how peak hype will be over earlier! Probably end of next week ;)
Replying to @rorcde
Really? In my daughter’s school the teachers started to give less homework and do more exercises in class after they realized students have been using chatGPT a lot. I‘m telling you, it is already there.
Thread (2/2)
@rorcde My personal experience is also that the number of times I had non-tech people bring up chatGPT went from „never“ to „too friggin often“ in the past months :)
Replying to @paul_rietschka
Oh yeah, the US has it's own very specific problems right now with a scientific view and reality... 😓
1
Replying to @paul_rietschka
Yeah. Instead everyone is focussed on regulation now it seems.
1
Thread (2/2)
@paul_rietschka While I was at the university, we had to come up with all kinds of excuses why we needed to buy new hardware. They said the desktop PCs provided by the university were „good enough.“ We built our own compute cluster with because the main infrastructure group couldn‘t.
Replying to @paul_rietschka
I mean all this super foundational physics research is like the best kind of science research. And if you see the enormity of the thing it is clear that it‘ll be expensive. With AI, not so much.
1 View thread →
Replying to @paul_rietschka
💯 this! Well as we all know ironically this is what OpenAI originally claimed they would do. It seems they achieved their goal irrespective of all the late changes in management 😂 With the help from Meta, who might have had their focus elsewhere…
1
Thread (3/3)
@paul_rietschka So thanks, Google/Meta/OpenAI, for the hard work. Let‘s take it from here :) It also seems the public funding agencies really didn‘t see the opportunity here. They put millions in the large hadron collider but didn‘t see that similar investments were necessary for AI.
3
Thread (2/3)
@paul_rietschka But they are not. And AI research has shifted from universities to big tech companies because universities couldn’t afford the money to put into compute resources. And now it seems we‘ve come full circle.
Replying to @paul_rietschka
Yeah, taking a step back, it‘s interesting what happened over the past 10 years or so. If Google were a „normal“ research institute, this would be fine, taking tax payers money, doing research, publishing results to advance the state of the art blah blah
View thread →
Replying to @GergelyOrosz
@matteocollina Sorry if this had been said before, but they also got rid of their whole logistics endeavor. For me that at least justifies the drastic cuts a bit.
1
Replying to @paul_rietschka
At the current pace of events being the best of „all week“ is already high praise 😅
Replying to @rorcde
@fedistats Using sendgrid right now, but the non-free tier is immediately $20/mo… but it is enough for now.
Thread (4/4)
@leonpalafox But I think it's limiting to look at smartness alone. What hold the human fabric together is connection and empathy. This seems to be something that's not even in the picture, with everything being focussed on input-output precision and so on.
Thread (3/4)
@leonpalafox I think there was also some discussion whether a supersmart AI would look upon us like we look on animals and not find humans worthy of preservation, and so on.
Thread (2/4)
@leonpalafox The discussion is along the lines of "smartness". What if the AI is way smarter than humans are, can we still control it, etc. But I mean what does smart even mean? I think there is some sci-fi picture in the back of a "God like AI" that steers the fate of humanities.
Replying to @balazskegl
Yeah, I also think there is a real concern here that needs to be discussed. But currently, I cannot shake the feeling the media is farming this for clicks, trying to outdo themselves with the most outrageous deadline.
1
Thread (3/3)
He had a nice office with a massive curved window front and when I said "You got the nice corner office" he just said "Oh, I got the office where there are no walls to put bookshelves."
3
Thread (2/3)
For the record, I met Geoff Hinton once when I was in Toronto for a week to apply for a PostDoc position. He was super nice, very approachable, and showed me the encoder stuff he was just working on, and his enthusiasm was contagious.
1
For future reference, a thread of people voicing concern about the impact of AI. To be clear, I'm not saying AI is safe. On the contrary. But I can't shake the feeling people are exploiting this for the clicks. <sarcasm>
3 View thread →
Posted this 10 days ago, cannot believe this was last week. Since then there was contract work, a comment by a friend on wtf I am doing, some less than positive discussions on mastodon re: opt-in, and I haven’t worked on it since that day. Quite the rollercoaster. 😅 twitter.com/mikiobraun/sta…
2
Replying to @boydroid
1. generate as much FOMO as possible 2. … 3. profit! 💵💶💴
1
The great thing about the M1 MBP is that it is so fast I didn‘t realize I need secondary indices for fedistats because all the data fit into memory and adding an index didn‘t really make a difference. But on a puny 2 vCPU/2GB VM it really really made a difference 😅
2
RT @rasbt: Losing track of all the recent large language model (LLM) developments? Here's is another comprehensive LLM model zoo, availab…
Re: Hinton, this whole discussion of smartness and control shows academia’s obsession with intellectual superiority. What about empathy and caring for others? If my life is controlled by something, I want to know that it understands and cares about the impact of its policies.
1 12
RT @swyx: LLM datasets be like: • First you start with CommonCrawl • Then you add C4, which is just CommonCrawl again, but dont worry abo…
Replying to @paul_rietschka
Yeah, one thing I see happening is that there's a bit more competition and options. And everything that's been true for dealing with data defined products still holds. People shouldn't believe the hype that "AI is solved."
3
Replying to @rorcde
Yeah, I think we're getting there... . Many open source models being released. We'll have to see about the quality, but it's not just OpenAI anymore. I feels like we're a bit past the big hype wave... .
Replying to @fluchtpunkt
Just so much, really 😅 I found that my @fedistats news feed captures a lot of interesting news articles (link below). Apart from that @rasbt has a great series of articles on substack with long lists of papers to read.
1 View thread →
I'm working on a presentation for a client on getting started with LLMs and generative AI, and reviewing the latest developments, the rate of new announcements is just mindboggling. Half of the stuff I'm talking about happened in the last two weeks.
2 16
Replying to @josipK
Yeah, social networks should be like new venues (like cafes, bars, parks, libraries, etc.) but when you join a new one it should be like you‘re scanning the room and quickly find people you know.
1
Replying to @MaryMacCarthy
@paul_rietschka Well, I just joined two days ago, so it‘ll be a while for me 😓
1
Replying to @MaryMacCarthy
@paul_rietschka Word on the street is one new invite code per two weeks. You in yet?
1
Replying to @fs111
Yeah if there is one company i would call grundsolide, it is Hetzner :)
2
Replying to @fs111
Yeah, Dropbox was also my first cloud storage, easy to use, multiplatform, etc. I never used the more advanced features and then migrated to Google Drive at some point. Also they were much more expensive at some point, I don't know whether they changed that.
Replying to @IgorBrigadir
Oh no, I'm not on the dev discord! I'm Missing Out! (thanks for the link!)
1
I thought the FOMO would be over once I joined BlueSky but now everybody is talking about building stuff on top of their protocol and I don‘t know where the docs are because the docs I find DON‘T SEEM TO BE THE DOCS EVERYBODY IS TALKING ABOUT.
1
RT @MaryMacCarthy: @paul_rietschka despite being the coolest of Tech Bros, I regret to inform you that I HAVE NOT YET RECEIVED A BLUESKY IN…
Replying to @paul_rietschka
I visited their office more than ten years ago. It was massive already back then. When I asked them why they needed so many people, they said, they're working on some big ideas... It seems they never really materialized...
1
Thread (5/5)
But if federation worked more in the way that you keep your identity but you can choose to be on a service fee/ad/data insights/premium content based model, we could maybe figure out a way to make all of this work without pouring millions of VC money into it.
Thread (4/5)
There have been other attempts, for making users pay for prime content, selling data insights, etc. Without federation, the problem is that you cannot experiment at all, because you need to bootstrap a whole userbase first, every time.
Thread (3/5)
So obviously nobody wants to pay for it, which means that the customers become the product and ads are the way. But nobody likes ads. Musk is now trying subscription models for elevated reach & features, API access, and organizations. But yeah... .
Thread (2/5)
Because let's be honest, we still haven't figured out how to make this work. Elon Musk is pouring enormous amounts of money into Twitter to keep it alive, and tries everything to break even, but essentially we haven't figured out how to make it work.
When it comes to federation, people mostly talk about freedom of speech and protection from censorship, but in this blog post from 2012, I make the point that federation is also an opportunity to explore different business models. blog.mikiobraun.de/2012/10/pheed-…
Media
Pheed, Tent.io, and the Future of Social Networks
Pheed made the news lately because they managed to get a large number of celebs to sign up for their service. Featurewise it’s somewhere between...
blog.mikiobraun.de
1 View thread →
I always say that the real danger from AI comes from hooking them up to systems that can do damage in the real world. When you worry about extinction-by-AI, that‘s how it will come about. Something along those lines vice.com/en/article/qjv…
Palantir Demos AI to Fight Wars But Says It Will Be Totally Ethical Don’t Worry About It
The company says its Artificial Intelligence Platform will integrate AI into military decision making in a legal and ethical way.
www.vice.com
4 11
Replying to @peterjliu
Yep, I agree. The real ingenuity was realizing input/output could be one stream to be completed.
Replying to @dagorenouf
What you think is a gap was the only time I felt professionally fully alive!
2
While you're all paying $8/mo for Twitter Blue, I got the early access to Twitter Diamond which costs $25k/mo!
1
Replying to @Quasilocal
Yeah, academia is harsh. Thinking back to the person I was… extremely competitive and insecure. It‘s always about „are they better than I am/are they worse“ and who is more successful. I would dismiss talks on the opening slide already. I don‘t miss it at all.
2
Replying to @Quasilocal
I think he could have upped the ego game by replacing „accept“ with „admit.“ THAT‘s how you demonstrate intellectual superiority!
1
Replying to @paul_rietschka
You know how we are. „Let‘s wait till this takes off.“ 🙄
Replying to @wrede
@dagorenouf Wait a second... wasn't relativity theory his sideproject? Wasn't he working at the patent office?
1
Replying to @wrede
@dagorenouf Daily outfit recommendation. He already tried to hack a solution by always wearing the same outfit. But maybe he was just constrained by the technology of his time!
Replying to @paul_rietschka
I was wondering for second if sama is who I think it is but that‘s actually his Twitter handle 😅
1
Thread (5/5)
So at this point it seems that the real experiment with building social networks is not the feature set (more or less converged), but what kind of cultural and values it should be built on.
Thread (3/5)
And mastodon, yeah. Just my personal impression, but there is an aspect of mastodon where some people are very sensitive about possible abuse that discussions sometimes end up being very black and white and focus on worst case outcomes.
One thing I noticed is that the "vibe" on the different social networks is quite different. People who are just joining Bluesky feel pretty elated, having just gotten off Twitter. And Twitter, well, you're all here, so you know... :)
2 View thread →
Thread (4/4)
But then again, if it's just the same bunch of people, why switch at all? Bluesky feels also so similar to Twitter. I mean if this were mastodon, it's probably just a new node.
1
Thread (3/4)
For mastodon, people built tools that help you find followers from Twitter again, but with the API situation at Twitter becoming worse, that's probably not a good path forward.
2
Thread (2/4)
I guess that's a bit different for everyone, but for me a social network needs to have enough activity that there's something interesting to check back, and also some people I know and interact with.
1
So I got the Bluesky invite and joined yesterday and immediately was reminded of last fall when people where flocking to Mastodon. The first question is "who to follow." and it's amazing that we haven't figured that out yet.
4 View thread →
Can't believe they make you download Edge to try out bing chat. The browser wars are so 1990s!
2
Replying to @rorcde
If you‘re not FOMOig over bluesky by now you‘re probably immune, Rodrigo!
1
Replying to @paul_rietschka
Maybe, but „the paradox of choice“ and „decision paralysis“ because of too many choices was stuff the product people at Zalando discussed non-ironically, so it‘s not just you!
Replying to @alung
Yeah I see… thanks for the insights. I think I‘m mostly using the Following tab so I see the tweets and languages of the people I follow. It‘s probably more useful for the algorithmic feed. I saw it had a „show this less often“ option, which I clicked. Let‘s see :)
Replying to @alung
Is my account so old, it has no language preference attached? Do people normally have preferred languages set?
Replying to @paul_rietschka
After decades of e-commerce one would expect that stuff is more optimized for immediate purchase decision making!
Replying to @paul_rietschka
Uuuuh, the M2 Mac Minis also sound very exciting 😅 But monitors, yeah... Too much choice!
Thread (3/3)
@paul_rietschka So yeah, the thinking goes "how about a less powerful machine that is more portable. But that's exactly what your Intel MBA is." And so on... .
Replying to @paul_rietschka
I know I'm sometimes critical of Google as a product company, but putting atomic clocks into racks is the kind of technological problem solving I wouldn't expect from most other tech companies.
1
Replying to @shrutimkrishnan
@krishnanrohit And I want the GPUs to be voiced by Benedict Cumberbatch!
3
Replying to @krishnanrohit
First crypto, now AI, do you sometimes wonder whether secretly the GPUs are behind all this and are pushing humanity from one compute intensive technology to the next to feed their need for raw electricity?
4
Thread (2/4)
The amazing thing is that NoSQL said that SQL couldn‘t scale horizontally because of the CAP theorem and then SQL said „hold my beer“ and not only did they lessen consistency requirements to scale (AWS Aurora etc.), they also used rack mounted atomic clocks to do both.
1
Ok, I‘m getting this for years. What‘s it‘s purpose? I‘m already getting tweets in more languages than I can read?!
Media
1
Thread (2/3)
For those of you who FOMO, right now as the US is sound asleep, and eastern Asia drifts into the evening, it‘s overwhelmingly Japanese people posting pictures of beer.
1
Thread (2/2)
@balazskegl Interesting, for older tweets, the views are not on the Twitter cards, but you still have these old style analytics pages, and it seems to be on a similar level. Well that at least…
1
Replying to @balazskegl
Very interesting, and unsettling. Views in general seem to be less than what they used to be. At least it feels like it. Hm. Do they have those view numbers also for older tweets? I should investigate…
1 View thread →
RT @timkmak: Good morning to readers. Kyiv remains in Ukrainian hands. Please RT this thread. It’s an important one. Some news: I’m lea…
Replying to @balazskegl
Yeah, in general talking to non-tech people, they don't understand how it all fits together and are often surprised already when they see targeted ads. They don't know about tracking pixels and how ad companies can see what you do across the web.
2
RT @fedistats: Just added a first version of hashtag trend classification. These should help to identify new events and filter out weekly h…
RT @fedistats: When you have a Ph.D. in machine learning but are too lazy to label some data points and run xgboost. https://t.co/2H3ys4WbG8
Media
RT @mlsec: Come join us in Berlin! Our graduate school is the perfect place to do research on machine learning, big data and, of course, th…
Replying to @InkmiHq
@KingOfCoders Ja... irgendwie kommt man doch immer wieder zu seinen Themen zurück... . Aber diesmal nicht realtime. Eine Erkenntnis war, dass sich der Aufwand nicht lohnt... . Dafür läuft ein Teil jetzt auf einem Raspberry Pi.
2 1
Replying to @balazskegl
Commenting here as well, so that the commenting doesn‘t distort the measurement.
1
Thread (2/2)
Joking aside, I think prompt engineering could be done with a good training set and something like genetic algorithms hooked up to a LLM in a systematic manner. In the end the prompt is just another hyperparameter, right?
1 4
Replying to @InkmiHq
@KingOfCoders For what it‘s worth, certain emojis that look like the „verified“ badge have always been forbidden, but yeah, they should just stop calling it „verified.“
1
Replying to @paul_rietschka
Yeah, this site, man… There is this one type of personality that cares zero about what other people think, and even if they are obviously wrong or it hurts other people, they just look back at you and say „but I want to do this“ and for some reason they‘re celebrated in tech.
View thread →
Replying to @paul_rietschka
Racking my brain what the most telling replacement reading for „AI“ in those tweets is. „Our new alien overlords?“
RT @mikiobraun: @ridtalkstech @halleberry Yeah, so you're "verified" if... - you're paying 8$ a month - were legacy verified and have more…
Replying to @ridtalkstech
@halleberry Yeah, so you're "verified" if... - you're paying 8$ a month - were legacy verified and have more than 1M followers - annoyed Elon so much he is giving it to you in spite. - ... what did I miss?
1 3 View thread →
RT @mmitchell_ai: I replicated this (my screenshot below). Really great example of gender bias, for those of you who need a canonical examp…
Replying to @paul_rietschka
Yeah, that's the weird thing, it almost doesn't matter for the bottom line for them.
Thread (2/2)
@paul_rietschka And doing research or working towards the occasional hype worthy event like beating people at Go is something completely different from creating a product, something that Google just isn't that great at.
1
Replying to @paul_rietschka
Yeah I think it's not uncommon that really big orgs allow for competing teams and don't try to be too efficient. I've heard similar things from Amazon. That's probably fine if you want to experiment, but when you need to deliver, you need to streamline.
View thread →
Replying to @bocytko
Yeah, I can already imagine all kinds of issues in real world productions… thanks for the links!
1
Replying to @rorcde
Yeah but I think the problem is that it’s mostly monetization and not really verification anymore.
Replying to @rorcde
Just joking because this is one of Elon‘s explanations. I don’t think it is related. I think the problem of the old way with getting verified was that the process was very intransparent and it wasn’t clear when you could get it or not.
RT @fedistats: I'm building a new feature that follows certain topics and automatically sends out new content that made it in to the daily…
Thread (4/4)
So anyone who worked with chatGPT, are they providing services for that? Because from the docs it looks like „it just works“ and I find it hard to believe that.
1
Thread (3/4)
Because if I learned one thing it‘s that eye balling a solution doesn‘t work. In „classical“ code you can test edge cases and you‘ll know „it works“, but with ML you really need good coverage and out of sample testing.
1
Thread (2/4)
Now having worked with ML based systems for the past decade or so, I‘m directly wondering how to evaluate this properly and make sure it works as intended.
Replying to @paul_rietschka
No idea. But to be honest I didn‘t know Google Brain still existed and interpreting the pre/post „merger of equals“ nomenclature I‘m not sure they still exist.
Thread (2/2)
@leop___ @j_bertolotti @Quasilocal But then they invented/created/discovered complex numbers, an extension of real numbers and there you have a square root of -1. But the strict mathematical proofs are independent of time.
Replying to @leop___
@j_bertolotti @Quasilocal I think the question is what „have a solution“ means. If the question is „does -1 have a square root“ this leaves open in which context. In the real numbers, it is true and will always be true that -1 does not have a square root.
2 View thread →
Remember, people who lost their legacy verified today, this is all to get a hold of the rampant bot problem!
2
Replying to @Quasilocal
Yeah, and also that you can prove that solutions exist without being able to say what these solutions are. Ah the beauty of non-constructive proofs :)
1
Replying to @balazskegl
Yeah, I definitely agree with you in terms of where the next step will come from. But I think there is still a lot of opportunity for others in terms of improving it as a product, or making it more commoditized, more effective, etc. And for Altman's motives, can't know for sure.
1
Replying to @ridtalkstech
Yeah, I‘m with you, definitely a 420 joke, maybe it is going to happen, maybe not. Wouldn‘t be the first time they went back on their word. Would probably even be the first time they didn‘t? 🤔
1 View thread →
Replying to @Quasilocal
Not gonna lie, proving that solutions exist with certain regularity properties for special kinds of equations was my then partner‘s whole master thesis at the time.
10
Replying to @dvassallo
(subscriber) Only got labels once while being in the US. Miss my (🌶️ hot takes)!
RT @mikiobraun: Is it just me, or has Sam Altman claiming that the age of big models is over strong "well, we struck a literal gold mine, b…
RT @fedistats: Another post on some basic usages and how to find interesting stuff on fedistats. Did you know you can look at trends by si…
Replying to @samuel_wong_
Aren't there some Google glass devices laying around somewhere?
1
RT @EwanToo: @mikiobraun 100% the same feeling, the era is "over" in that they've got a good thing going on with their duopoly with Google…
Thread (2/2)
At the same time, I'm convinced that we need smaller, more energy efficient models, and I'm sure we'll get there. So he's not wrong probably, but it just feels soooo... sketchy.
4
Is it just me, or has Sam Altman claiming that the age of big models is over strong "well, we struck a literal gold mine, but I wouldn't recommend this spot to anyone else" energy?
1 12 View thread →
Replying to @InkmiHq
@KingOfCoders Some people are doing everything to be in the news these days.
1
Sundar Pichai making the link between AI safety and nuclear arms treaties is just seriously bad taste.
2
Gave chatGPT a try to do some summarization. Would you rather get the original list or chatGPT's summary in your inbox?
Media
1
RT @IgorBrigadir: Retweeting again to solicit more responses. Especially now for journalists. Clearly as a field, we are failing at outreac…
Thread (2/3)
The difference however is that the transition from short term memory to longer term memory takes place automatically. You read an article and still remember it the next day, but with LLMs you need to be quite explicit about all this.
Keep thinking about the different memory levels in LLMs. There is the low level "muscle memory" that takes ages to train, "finetuning" that you can update with acceptable effort, and the short-term memory of explicitly provided history tokens. Not so different from human brains.
1 3 View thread →
Replying to @fedistats
That flight tracker is also trending under #SpaceX... well well... 😓
Media
Replying to @InkmiHq
@KingOfCoders Yeah I was also hunting for something that gave a bit more resistance than usual Cherry Browns. Ended up with Glorious Panda switches. They also claim the position of the bump is a bit earlier. I don't know. But I like it.
1
RT @fedistats: Working on a new feature: Topics with notifications. So using fedistats myself I realized that the daily trends are nice, bu…
@BenHanowell Yeah, ever since this hit the mainstream it feels like no occasion is safe anymore from the „chajeepeetee.“
Replying to @paul_rietschka
@verge „Site owner now also builds cars: what to expect“ 🤣
1
Replying to @paul_rietschka
@verge That bit sounds still somewhat US centric. China is a such a massive self-contained market, and times are past where they are just manufacturing.
1
@gsingers @tunguz I don‘t know… maybe if you call it OpenSVM? That being said IIRC there were a few SVM based startups back in the day.
I might be biased but having your bot send automated direct messages with latest #ai articles to you is kewl. Keep at it, little bot, you‘re doing great! 🤖
Media
1 View thread →
Thread (2/3)
@tunguz On a less sarcastic note, I've interviewed hundreds of scientists since 2015 and could really see a shift from SVMs to xgboost to deep learning over the years for essentially the same toy problem I used. It pays of to know what to use when, even in the days of "AGI."
2
No time like a late Friday afternoon to push to prod and fix a missing dependency while looking at the clock to see if I can make it before the next batch job is starting.
3 View thread →
Replying to @filiphracek
@year_progress Man, this is sad! @year_progress was my constant companion through the last years, reminding me of time’s inevitable passing and more than once I checked it to know when it‘s New Year in UTC land.
2 17
Really, God-like AI?!? I said for years that the first thing I would want is dog level AI: - bonds with us humans 🙋🏻‍♂️ - can fetch a frisbee out of air 🥏 - notices and cares how we feel 💕 - follows commands 🗣️ - belly rubs! 🐶 Nuff said! Life is ruff!
2 10
@BenHanowell @paul_rietschka Definitely! Sorry, I couldn’t resist to try a witty reply here.
Replying to @paul_rietschka
@BenHanowell You laugh, but once we have enough AIs making business decisions, this will all totally make sense.
1
Replying to @paul_rietschka
Saw a meme today that said: „dear autocorrect, it is NEVER duck“ next to a rubber duck. That was fair use imho.
1
Replying to @InkmiHq
@KingOfCoders For a second I thought this is a new rectangular keyboard layout. (Be still, my heart… 😌)
1
Replying to @AdamSinger
Strong Apple saying their GPUs are more energy efficient, well, energy. At least their chart had labels.
Media
4
RT @iamharaldur: The future was supposed to free up our time from work so we could spend it on higher pursuits. Instead we created AI tha…
Replying to @wrede
Oh yeah, I do hate those "isn't X just like Y?" question with a vengeance. That's just lazy thinking. Do your research and form an opinion of your own. It's worth your time!
Replying to @breuderink
Well I was exaggerating a bit for the sake of Twitter, but you can write matrix multiplications as SQL joins, yes 😅
RT @channingwalton: @nraychaudhuri It doesn’t need a title. It’s like saying a person using excel is Function Engineering, or Word is Par…
Thread (2/2)
@ridtalkstech I‘m also pretty sure I once read about a human who failed the Turing test, and he‘s written a book about it, but I can‘t find it anymore.
1
Replying to @paul_rietschka
That „paper“ and the one about jobs „endangered by AI“ are pure form research paper styled marketing are what they are.
1
Thread (2/2)
SQL, map reduce, matrix multiplication, for loops, at some point you‘ll understand that they are all embodiments of the same universal concepts. Don‘t let the AIs take that away from you! 🙇🏻‍♂️🙏
4
Replying to @GergelyOrosz
And you didn‘t even mention their name, so yeah, probably not the most impactful marketing effort 😅
I honestly wasn‘t aware Twitter Circles was a new feature?! Ignoring the buggy implementation, hasn‘t this been tried several times and never really kicked off?
1
RT @mmitchell_ai: Finally have a moment to read MSR's "Sparks of AGI" paper. I'm going to do something very crazy and *live tweet* my thoug…
RT @fedistats: Hmmm… some nodes are definitely lying about their number of users and posts… 😅 https://t.co/153kf0rPsp
Media
RT @mikiobraun: Trends on the fediverse show quite rich behavior. There are periodic topics like #press, or #photography, then there are ev…
Replying to @leonpalafox
@easyNotComplex @GergelyOrosz Yeah I agree. We probably have more "trust" in Google that they will exploit it to their own benefit but not accidentally leak all those internal memos you asked chatGPT to summarize to the public.
2
Thread (3/3)
@easyNotComplex @leonpalafox @GergelyOrosz For example, a model that had your data in there among billions of other data points, is that sufficiently depersonalized? Normally you‘d say yes, but because the systems also do information retrieval, can you make sure the specific data won‘t come up again? And so on.
1
Thread (2/3)
@easyNotComplex @leonpalafox @GergelyOrosz They don‘t seem to ignore existing laws, but I think the problem is that for a system as complex as theirs, it is really hard to tell when and whether your data has been sufficiently depersonalized, aggregated, and so on.
Replying to @easyNotComplex
@leonpalafox @GergelyOrosz Depending on where you live, you can request to have your data deleted, per their privacy policy (that I had mentioned above)
Media
1 View thread →
Replying to @GergelyOrosz
It is pretty save to assume that they collect all the data, use it to monitor and evaluate performance, use it to further improve models, and so on. That‘s just standard procedure for building data driven products. And yes, GDPR is to give people control for exactly this.
3 View thread →
Replying to @GergelyOrosz
@ogbonigwe1 @Carnage4Life Yeah, my impression from stories in Germany is that you can lay off people but it‘ll cost you. I guess many don‘t want to keep the job but want to get better settlement. It might sound weird to US people, but in a way it is very capitalistic, but different.
27 View thread →
Replying to @paul_rietschka
@mmitchell_ai While having lawyers in court arguing that there is no fraudulent behavior regarding dogecoin no less. It‘s all a joke to 🍉.
1
Replying to @paul_rietschka
Yeah, turns out you can pack a lot of compute into a big hall. Same for Amazon distribution centers. Sure, they are big, but compared to the area they serve it's a bit mind boggling.
1
Replying to @paul_rietschka
I still can't wrap my head around the fact that there are places like Ashburn, Virginia which host a cluster of data centers.
1
Replying to @FlaxSearch
In Horizon: Zero Dawn, I think the AIs are equally male and female. Then again, they are more programs in a big machine controlling all those robots and not really that humanlike.
RT @paul_rietschka: @mikiobraun By far the most realistic-feeling science fiction re: AI is Asher’s polity series. The idea of offloading r…
RT @mikiobraun: Required reading for AI, category fiction, a primer. Order roughly from "caring" to "let's get over with humanity already"…
RT @fedistats: Sigh, #dogecoin is trending on the fediverse, of course... But some good links there if you want to read all about it... ht…
Required reading for AI, category fiction, a primer. Order roughly from "caring" to "let's get over with humanity already". Massive *spoiler alerts* => 🧵
1 3 View thread →
RT @dosinga: Spreadsheets, Python and LLMs just work so well together, it still blows my mind
Replying to @InkmiHq
@KingOfCoders No idea… the film industry? Btw, whenever I want to make some minor pictures edits, GIMP is still my tool of choice!
Replying to @madsh
@paul_rietschka Ah man. I‘m not following the discussion, but that‘s not good. Haven‘t people read Asimov?!
Replying to @madsh
@paul_rietschka And yes, I agree! There is talk about AI regulation but the very first rule should be that AIs must always self identify as such.
Replying to @madsh
@paul_rietschka You haven‘t seen Westworld? It was a TV show where we have robots that look so real they are almost indistinguishable from humans but held in sort of an amusement park till everything goes horribly wrong. Pretty good.
Thread (2/2)
@paul_rietschka There was a story at Zalando that when they launched in Italy, people would call the customer service just to check whether they are real. We need something like that for AIs.
Replying to @paul_rietschka
Decades of electronic communication have definitely primed people to connect with light signals on screens. Ok, maybe this goes even further back. I think writing and letters are the real culprit here! :)
View thread →
Replying to @paul_rietschka
People are making up new interpretations so quickly, there is no way to keep up. But yeah, the fundamental mistake is that there is something humanlike inside.
1 View thread →
Replying to @paul_rietschka
Yeah, I find it pretty tiring trying to refute all those things people put into LLMs. People say LLMs hallucinate, or lie, or are biased, or are overly confident etc. Whatever happened to Occam‘s razor?!
2
RT @leonpalafox: When you order Uber in Mexico, sounds pretty religious https://t.co/PZO8aMqGrN
Media
RT @fedistats: A first post on the what and why of fedistats and some insights into the fediverse you can get from there. https://t.co/k3V…
Replying to @bobbruno70
Yeah, true… the vibe has definitely changed. Maybe computers have also become so much more part of out daily lives?
Thread (8/8)
I mean let‘s also spin this further… if we have a considerable amount of AI generated code in the future, more code that nobody understands, but you will eventually have to look at when the AI tools stop working... Sounds like you‘ll need a lot of expertise going forward.
1
Thread (7/8)
I want to add that people questioned why I wanted to study CS back in 1994 because „these computers will soon program themselves.“ From all I can see, for those who write code, things got increasingly more complicated, not less.
4
Thread (6/8)
We add complexity to our libraries and APIs, we use language that don‘t allow us to properly abstract all that. Patching this up with multi-billion neuron sized smart autosuggest is one way to go at it, but it is covering up this problem.
Thread (5/8)
There is a side here that‘s seldom mentioned which is why we ended up in a situation where we‘re in need of such tools. The real problem is that we‘re often required to memorize dozens of lines of boilerplate code to get something done.
1
Thread (4/8)
Sure, we‘re getting a new bunch of tools that are amazing and are changing programming the way stackoverflow changed it, but that is just a part of all the activity and skill that is required to build software.
Thread (3/8)
Coding is not just about churning code that works, it is about designing, sometimes discovering, the right abstractions, building structures and tools along the way.
2
Thread (2/8)
Also given the current state of the art, it takes real expertise to spot the mistakes in code that superficially looks right, but fails for subtle reasons.
2
Man if I see one more „the end of programming is nigh“-sayers… programming and creating large scale systems that are resilient, architectured well, maintainable, that‘s so much more than „how do I OAuth in typescript.“
2 11 View thread →
RT @fedistats: This is going to be interesting (also in terms of observing the hashtag…). Fedistat gets updated once per hour and it just h…
Replying to @francoisfleuret
No, I even found and installed the update, but it‘s still keeping the lane. Maybe I‘m in a different A/B test group? My account is weird sometimes…
Replying to @francoisfleuret
For me it only goes to the top. Maybe I still have the old version…
@gsingers Yeah, don‘t get me started! They also make it sound like we can only go further in OR work on guidelines. Don‘t we have gazillion of people working in that field now?
Did they lower the resistance or whatever that‘s called in the Twitter app to make it easier to accidentally switch between the Following and For You tabs? Happens to me all the time today. Can‘t remember it was like that before.
1
Mi: Berlin -> Köln, Stellwerksstörung, Zug soll über D‘dorf umgeleitet werden. Bis der neue Fahrplan da ist, ist die Störung beseitigt +1h. Do zurück: ICE defekt, fährt nur 160 +40m… 🙄 Läuft
1
@chjrwis @krishnanrohit Yes, and taking a step back. But there is something deeply psychological about catastrophic outcomes that makes it hard to not disregard the probability. You tell yourself it is unlikely, but IF it happened it would still be devastating, so the prob doesn‘t matter.
1
Thread (2/2)
There are few cases when this can be exploited for PR, but yeah, it seems right wing politics and getting everybody‘s attention about the AI product you‘re working on are two real world examples.
3
Just try to keep in mind that we‘re bad at dealing with highly improbable but devastating outcomes. When you make the switch from „ah well, that probably will never happen“ to „but what IF it happens,“ they got you. This is exactly what the AI hype is feeding you right now.
2 9 View thread →
Replying to @krishnanrohit
Awesome read! I see this was from (gasp) three months ago. Did anything change for you since, let‘s say, two weeks ago? ;)
1
RT @krishnanrohit: Can we please all of us read this Time article and agree once and for all that the emperor has no clothes? All biologic…
Replying to @krishnanrohit
Unless we let it! I always held that the real danger is from giving these algorithms too much power, let them decide things with real world consequences, or hook them up to systems without proper oversight.
1
Replying to @rasbt
When DeepBlue beat Kasparov, I was in my second year of studying CS. The efficient algorithms professor said during the lecture: „have you seen this? It is over! The machines won!“
2
What is the state of the LLaMa model that's floating around? Is it true it is not public yet but has been leaked? If true, what about projects like dalai that let you easily download and run the model and which is hosted on github, owned by MS, who has a big stake in OpenAI?
1 1
Replying to @gmarkb
@anildash @andishehnouraee Shameless plug here, I'm working on a site that let's you explore nodes and trends on the fediverse to help you discover great content => @fedistats
2
Whoa, first time I‘m seeing one of these. Which stage of the Gartner hype cycle is this again?!
Media
5
Replying to @paul_rietschka
@timnitGebru Me: who is this Chris Murphy guy? Oh, a US senator. 😰
4
Weird thing, I got a bot saying Andy Karpathy has unfollowed me (not @ any of them). Now I am wondering, too much anti-AI sentiment?! Too much @fedistats promotion? Or am I just not that funny? 😅 It must also be weird to have a bot publicize when you unfollow people!
1 View thread →
Replying to @paul_rietschka
Yeah, it is still supervised learning. In a way we‘ve come far but still is how good you can be on out of sample input, output pairs.
1
Replying to @roydanroy
Can someone hook that up with some grant database, then we can train models on grant-to-latex tasks!
RT @beeonaposy: Every time I read about how people are using AI to automate writing SQL queries, I think of this tweet.
I can only recommend looking at alpaca.cpp and talk to an LLM in its pure form. It becomes very clear it is just freely associating.
1 5
Replying to @paul_rietschka
Wasn‘t that the point of meta’s llama? Stanford‘s alpaca say they were able to finetune a thing that follows instructions for a couple of compute hours. Which finally brings us back to where we were in terms of training-your-own-models.
RT @killedbygoogle: Aaaaaaand...Bard has sung its last song. The LLM Bard existed from March 21, 2023 - March 23, 2023. https://t.co/Wz26…
Replying to @wrede
Um, no, but pushing changes to prod when the weekend has already started.
Replying to @rorcde
Ha, research legacy! For me personally, kernels are still relevant for specific problems, where they compete more with xgboost and friends. Although I met someone recently who said string kernels still work better for their application area.
Replying to @paul_rietschka
Next thing you know, someone made an SVM with a string kernel work!
Alright, I'll give in. Time to read some ML papers... . But if I'm doing this, let me at least do it at my most cynic, sarcastic, thinly veiled envious - just like a real academia PostDoc.
1 3
Thread (2/2)
@rorcde I have to admit I find this kind of data always very interesting. Somehow tapping into the stream of thousands of people and extracting trends and patterns. After mastodon really kicked off last fall I wondered whether it had reached critical mass, and I think it did.
1
Replying to @rorcde
Well, yeah... so I made a list of ideas, and then picked the one that seemed most fit to start, part because I did something like it before. But the tech is pretty different. I went more old school data pipelines and postgres instead of any fancy approximative counting :)
1 View thread →
Replying to @dirkriehle
@owehrens Sind auch nur Menschen :) Was zuletzt ein paar Mal geklappt hat, war morgens zwischen 7 und 8 Uhr einen Termin für denselben Tag zu finden. Gut, dass das ging, aber dass quasi immer alles 3 Monate im Voraus schon ausgebucht ist, ist auch nicht gut…
1
Replying to @owehrens
@dirkriehle Ich habe es mal gewagt EINEN Termin zu machen für die Reisepässe meiner beiden Kinder. Mir wurde gesagt, ich hätte zwei Termin in Folge buchen sollen. Ich weiss gar nicht, wie das hätte gehen sollen über die Webseite.
RT @paul_rietschka: @mikiobraun Yeah, there are some real changes coming to tech, spec. around code generation. Forget low code, we’re movi…
Replying to @bocytko
Yeah, being fascinated by your own demise is not everybody‘s cup of tea…
Replying to @paul_rietschka
Yeah, I barely get past the headlines without all my BS detectors going 🚨🚨🚨
2
Replying to @paul_rietschka
They are sure trying to talk themselves into making AGI happen. A low point for science.
1
One side effect of the current AI hype is that mainstream people are being exposed to academic writing... Why didn't that happen in the pandemic?
3
Thread (7/8)
And at the same time, we extracted "common sense" that previously took decades to assemble manually. This is remarkable, although there are still more challenges to take. But that's for another thread :)
Thread (6/8)
It seems we've achieved the same thing using large language models. Essentially by learning the sequential structure of data, we have somehow created a machine that generates that intermediary representation, that we don't understand but that seems to be good enough.
Thread (4/8)
Machine translation used to be the hunt for an universal grammar and parsers/generators as intermediary representation to translate between languages. As you can image, given all the quirks of natural language, this was difficult.
Thread (3/8)
Started in 1984, it is still ongoing for almost 40 years. Painstakingly trying to distill "common sense" so that machines can reason about reality. I see parallels to the way ML has completely changed machine translation here.
Replying to @gridinoc
@killedbygoogle The wound that was Wave runs deep! So much ahead of its time. Almost as deep as Google Reader!
2
Replying to @bocytko
Ah, interesting idea. The way it is right now, most of fedistats data processing is in a pipeline that stores the results in a postgres database. I guess one could run streamlit against that 🤔 But I'm also not that concerned with how it looks tbh.
View thread →
Thread (2/2)
He also offered to make a marketing video, but wants to do it in the style of Honest Reviews. "You like websites that are fast, bug free and have great design? Well then this isn't for you!" 😂
4
RT @fedistats: So AI is pretty well covered, but if you're into cryptocurrencies, #crypto, #cryptonews, or #bitcoin look like good starting…
Replying to @rorcde
Oh yes! It is an insane hype bubble right now. I think most if not all of those articles that do predictions on which jobs will be eliminated are securely within that bubble.
1 View thread →
Replying to @rorcde
Well, skimming the articles people aren't that impressed. But feel free to follow the links trending on @fedistats right now :)
RT @fedistats: I got tired of doing screen shots to share interesting trends, so I added proper open graph and Twitter card metadata: http…
RT @fedistats: Whoops, when you're doing that quick deploy to try something in production. https://t.co/z2JsOf2g7s
Media
Replying to @krishnanrohit
Ah sorry, mElon is maybe only how I call him. I was joking whether Twitter Blue makes you force follow Elon Musk, but probably not. ;)
1
Replying to @bocytko
You mean apart from @fedistats? Joking aside, yeah I‘ve seen some pages/github repos tracking models and also open source clones but can‘t remember right now.
View thread →
Replying to @CFDevelop
That there is that OpenAI „research paper“ discussing job security people is the latest escalation of research style going marketing.
1
Replying to @mtantawy
It‘s not even correct, it should read „Essen im Bahnhof“! Infuriating!!
2
Replying to @krishnanrohit
Ridiculous. This is like half of the possible app icons for IceCubes
Media
3
Thread (2/3)
There is general information about nodes, the languages, and hashtags they use to explore the different offerings. And there are daily trends based on toots for hashtags and sites. Go off and explore!
RT @lemonodor: GPT is so good at bullshitting that the replies are filled with people who, even after being told it can't request URLs, sti…
Replying to @leonpalafox
Next he‘s going to tell us we should give him money because only he can protect us from the AIs.
RT @mikiobraun: It's paradoxical, I've been interested in AI since my late teens and it was one of the main motivations to study computer s…
Thread (2/2)
@didou @catherinebuk Especially in the demos and the marketing… But why do they focus on text and images but not on empathy and social interactions? Good questions. Part of it is technical feasibility (text is abundant on the web), but labelled data for social interactions less so.
Replying to @didou
@catherinebuk Yeah, it seems to me that current technology goes down a path which started out as how research was formulated (especially how metrics were defined and the evaluation was set up). And then creativity ran free what one could do with that. But there is so much cherry picking.
View thread →
Replying to @paul_rietschka
Yeah I think that already starts to happen. Saw a post on LinkedIn (but can‘t find it anymore, of course) that listed several of them.
Thread (12/13)
Don't get me wrong, I've nothing against the money, but at the same time we're fed the narrative that this is about mankind and the next step of humanity blahblah. Let's just agree this is a gold mine and move on. abcnews.go.com/Technology/ope…
OpenAI CEO Sam Altman says AI will reshape society, acknowledges risks: 'A little bit scared of this'
The CEO behind the company that created ChatGPT believes artificial intelligence will reshape society as we know it, but admits there are...
abcnews.go.com
5
Thread (11/13)
If I asked you a year ago, would you have believed that Microsoft would beat Google to adding AI features to their office suite? Already there are dozens of startups scrambling to build products on top of this new technology.
3
Thread (10/13)
And lastly (3) the MONEY. By now it is pretty obvious that a lot of money can be made with AI. Remember that Microsoft invested $10bn into OpenAI? Did you think that Bing search would ever make headlines again? forbes.com/sites/qai/2023…
Microsoft Confirms Its $10 Billion Investment Into ChatGPT, Changing How Microsoft Competes With Google, Apple And Other Tech Giants
Microsoft finally confirmed that they would be extending the partnership with OpenAI, the company behind the new AI-powered ChatGPT tool. We look at...
www.forbes.com
2
Thread (5/13)
We've grown accustomed to this fragmented form of non-discourse, but it does avoid the question what this new thing really is and what the consequences are.
6
Thread (4/13)
Others say "I don't see why people are concerned about consciousness. Maybe we're also just biological computers." Which is also true, but then again, this touches really far reaching fundamental questions of humanity.
2
Thread (3/13)
That's definitely true, but then again, this piece of technology is different from everything that came before. Others focus only on the amazing things it can do, which it can, but then again, it's not about the amazing things but also about the impact and the consequences.
2
Thread (2/13)
(1) The way the discussion is fragmented. I think this is very much the fault of character limits in social media, but many people pick one aspect and stick to it. E.g. saying that every new technology (calculators, the newspaper, television) lead to people being concerned.
1
It's paradoxical, I've been interested in AI since my late teens and it was one of the main motivations to study computer science, and yet, the current state, especially of the discussion, infuriates me. I've tried to understand just why, and here are my main takes:
6 25 View thread →
Are you ready to use the new AI features in MS Office in your company? I'd say wait till Microsoft trusts their AI enough to let them handle customer service.
4
Replying to @paul_rietschka
We jest, but I already saw a post where someone claimed chatGPT wrote their website for a couple of dollars, while the freelancer wanted to have $5k!
RT @stefan_will: @mikiobraun Seems like everyone's trying really hard to speedrun the Gartner hype cycle.
Replying to @rorcde
One could also say you‘re on the very optimistic side :) But as I said, I was talking about the optics, so in fact about how it appears in the press.
1
Replying to @rorcde
I think we‘re making different assumptions here, and I find it a bit more worrying than you. What‘s really going on, who knows.
Thread (3/4)
@rorcde BTW, I'm not saying that I believe Microsoft is doing evil stuff or doesn't care about the impact of their AI systems. I just said the optics of "double down on AI" / "lay off ethics team" don't look good.
Replying to @rorcde
I think this is specifically one area where you wouldn't want to look at it from an ROI perspective, to be honest.
Thread (2/3)
I think it is about time, Atlassian ups their game and have an AI assistant to produce full project roadmaps, milestones and tickets. Oh, and status reports! Management needs those status reports!
2
Replying to @paul_rietschka
The thing that drives me nuts is I never end up in these situations where it is „piles of money, but…“ 😉
1
Replying to @paul_rietschka
I thought they started out with noble intentions but then couldn’t just say no to the attention (pun intended) and the piles of money that are up for grabs.
1
Replying to @paul_rietschka
@verge Eventually we‘ll have brain implants that take care of the whole „communication“ thingy.
1
Replying to @paul_rietschka
@verge There‘s already so much superfluous creation of text in the corporations, this will just add to that. Next step: an AI that boils AI generated text down to the gist.
1
Replying to @paul_rietschka
Ooookay, so what they‘re saying is „don‘t worry, we‘ve got our eyes on the prize and won‘t let ourselves be slowed down by those ethics people.“?! And yeah, I saw that paragraph where they said they wouldn‘t disclose any of the implementations details… very… „open“
1 View thread →
Replying to @wrede
@progress_season You mean I‘ll miss winter after the whole climate change?
Replying to @paul_rietschka
Yeah, meaningless text scaffolds. Sometimes I wonder just how many people have forgotten or never knew that text is about communication and connection and not just "I need a page of text."
1
Replying to @krishnanrohit
Sorry, didn‘t meant to get timeouted again, I was just wondering why Hemingway…
1
Replying to @clairikine
Monday: sneakers, Tuesday: rain boots, Wednesday: back to winter boots 😅
1
RT @mikiobraun: The people from Github's ReadME project were kind enough to ask me to write about the interplay between open source and mac…
The people from Github's ReadME project were kind enough to ask me to write about the interplay between open source and machine learning. Having been part of a group to promote OSS in ML more than 16 years ago, it was great revisiting this topic. github.com/readme/guides/…
Media
The unique origins of open source in machine learning
With the help of open source, machine learning has grown far beyond its academic roots. Understanding this history can help navigate the technology’s...
github.com
5 10
Logged into Xing to call one of the very persistent recruiters off. Saw that I had a dozen messages from recruiters. It was time, I finally deleted my Xing account.
1 7
Replying to @IgorBrigadir
@TwitterDev From my personal experience I don‘t think there are many use cases where you can make that much money from the data. I least I haven‘t found them 😅
1
Replying to @IgorBrigadir
@TwitterDev I just understood that this is not just a meme. The pricing seems to be based on „how much money do we need to make / how much API customers we have“ 🤯
4
Replying to @__AlexMonahan__
Alright, well that would be another explanation. Once encryption is broken, the only way is to put the data in a highly fortified place.
1
Replying to @paul_rietschka
I have to admit I'm unfamiliar with those works! Time to catch up!
Thread (2/2)
@paul_rietschka I found The Expanse much more believable with the handheld terminals, but then again that's Hard Science Fiction. And also an extension of our timeline whereas nobody knows where Star Wars is located.
Replying to @paul_rietschka
Yeah, now that you mention it, it IS very weird. Also this whole "we have holograms but yet we need to hide this memory chip in a stubborn little robot because we have no safe means of transmission."
View thread →
Replying to @paul_rietschka
Yes, and nobody realized yet that restricting physical access is insufficient from a security point of view!
Found this in my photo library from May 2017 and now I‘m wondering if this is really about the Death Star schematics or just a shitpost.
Media
7
RT @AdamSinger: the AI we should be rooting for / the AI a bunch of tech nerds are rooting for https://t.co/5q2upXoVdO
Media Media
@dovetailgames I recently got the NEC: New York - Trenton extension for my son who is a big train fan. But turns out, it is apparently very similar to an earlier DLC and quite buggy. It was a gift on @Xbox, so no refund possible. I hope they'll release an update!
View thread →
Thread (2/2)
@rorcde @CFDevelop For a full rewrite to work, you need people who know the current system inside out, have had their learnings, etc. And you need to decide what to cut, what you need, etc. and for that you also need people who are probably no longer employed at Twitter.
Replying to @rorcde
@CFDevelop Not a surprise at all, I think that's the natural course for all large scale software systems. Still, a full rewrite is also a very challenging thing to do, especially while you have another system still in production. If at all, gradually rewriting things one system at at ime.
2 View thread →
Replying to @rorcde
Yeah I wouldn't as well, but I'm starting to see people use it (or the bing chatbot) for exactly that.
1
Replying to @CFDevelop
Have you seen that verge article claiming that there was just a single engineer working on that project? The org is brittle is what it is!
4
Are you all still googling or are you already chatGPTing… heck we need a better term…
1
Replying to @francoisfleuret
My position has always been that the danger with any kind of "AI" lies in us putting too much trust into these things and hooking them up to systems that can do real damage.
1
RT @nytimesmusic: Breaking News: Wayne Shorter, the enigmatic, intrepid saxophonist who shaped the color and contour of modern jazz as one…
RT @wrede: Wow! These guys are using my product to raise funds for Ukrainian soldiers. 🤩
RT @thatgreygentlem: A conversation I just had: Me: “Hey are you around? I need help knowing my worth.” Friend: *calls me* “Listen here,…
Replying to @bobbruno70
Always remind me of Joschka Fischer who once said (if I remember correctly) „alright, I take full responsibility for what happened, so let‘s stop talking about it already!“
RT @random_walker: Considering that OpenAI did a decent job of filtering ChatGPT’s toxic outputs, it’s mystifying that Bing seemingly decid…
RT @jessetanderson: We have 63 responses for the 2023 Data Teams survey. I'd like to hit at least 80 for the analysis. We already see some…
Replying to @stephenroller
You mean to generate the answer? Yes, but do you feed the whole prompt again or do you do that once, and then start the system from that state at the beginning of a conversation? (At least that's my understanding of how these systems work)
Replying to @stefan_will
Very true :) Lazy product work, or ingenious "press release to product" hack?
1
RT @IgorBrigadir: Personally, I've settled on condemning this entire "Sydney Bing being a bitch" thing as "It's bad on purpose to make you…
Replying to @stefan_will
Yeah, saw that. First time I saw just how complex these prompts can be. Mind boggling.
Can I get a bit more engagement on my tweets or do I have to buy this platform first or what?
1 1
Replying to @IgorBrigadir
Yeah, you're right. It'd do the same just to mess with them. ;)
1
Thread (2/2)
@IgorBrigadir Maybe it's all PR and marketing at this point. "People aren't impressed by grammatically correct answers and extensive knowledge anymore, we need something more... spicy. Any ideas?" "We could make it emotionally unstable. I mean we can correct it later." "Make it so."
2
RT @gaspodethemad: And so it begins: the bingularity (I am sorry. That was terrible. I'm done) https://t.co/epaGpiMrX5
Media
Replying to @IgorBrigadir
You mean they couldn't resist and added "spice it up a bit. Look alive. Make them wonder whether you're real" to the prompt?
1
More stories about Bing's "emotional instability." It's actually an interesting question why chatGPT didn't show that too much. Do they actively apply sentiment analysis on the outputs?
1 3 View thread →
At this point 🍉 should have just created his own mastodon instance and patch the backend (the power of open source!) such that everyone sees his posts.
1
Thread (2/4)
And (b) Bing does not get depressed because it has no feelings. Stop reading emotions into these things. It is text that somewhere in the hidden state has some representation of a tone of sadness, and through text associations it may appear that there is some natural logic to it
A lot of interesting behavior of the Bing AI. But you need to keep in mind that (a) the amount of training data that is used is unfathomable. Probably somewhere in there was a very similar situation to what is being talked about right now. >
2 View thread →
Replying to @_deep_yearning
I blocked him a long time ago and haven’t regretted it once.
2
Replying to @mchmarny
@Twitter Oh I agree with you. I think it is one of those things where it was supposed to be good for something (higher engagement by more relevant, personalized content?) but now it is just there stubbornly because people say it should go and it doesn‘t want to.
1
Replying to @wrede
@TwitterDev Stark inability to read the room? Absolute clownery. But yeah, let‘s wait another couple of days.
1
RT @martingoodson: ChatGPT gives literally insane answers like this one and yet google bard gets incinerated for minor factual mistakes? ht…
Replying to @paul_rietschka
You mean like every time they open the box to fix it, a cat dies?
Replying to @shanselman
"Twitter Blue" ... "in the US" ... ?! also what is wrong with the Captialization?!
Replying to @paul_rietschka
Didn't Einstein say something about beaming information from point A to point B instantly?
Replying to @JamesHornick
^^This. So true! Where are the chatGPT conferences with Conversational AI hackathons and raffles? Are investors sleeping? We need to pour our funds into this!
1
Replying to @j_c_cabrejas
Yeah apparently new rules, there will be a free tier and new deadline is Feb 13… 🤦🏻‍♂️
Thread (2/2)
@bocytko Well, I mean if you decide to go down that path where your product is a black box and the only way to control what it is doing is by feeding it billions of examples, it‘s no surprise you cannot be sure what it does. 🤷🏻‍♂️
This is wild. Suddenly, after having an absolute niche search product for decades, Microsoft appears more competent than Google? And we‘ve seen chatGPT weasel its way out of inaccuracies for months but Google‘s BARD must be perfect? theguardian.com/technology/202…
Google AI chatbot Bard sends shares plummeting after it gives wrong answer
Chatbot Bard incorrectly said James Webb Space Telescope was first to take pictures of planet outside Earth’s solar system
www.theguardian.com
6
Replying to @progress_season
No idea what they are doing. They either cut access without warning over night... or announce cutting it off, but then it's unclear when or how... ah well...
View thread →
Replying to @timserpas
@year_progress @NicCagePlotBot @RikerGoogling @emotionlocation @swear_trek @progress_season @StealthMountain Stealth mountain was a legend.
2
Replying to @swear_trek
@year_progress @NicCagePlotBot @RikerGoogling @emotionlocation @progress_season @StealthMountain Yess
1
Replying to @swear_trek
@year_progress @NicCagePlotBot @RikerGoogling @emotionlocation @progress_season @StealthMountain Ah, my bad 😅 So the good thing is you'll stick around, right?
2
Second day in a row where I turned on the coffee machine but forgot to put in the coffee. It‘s going great.
6
Replying to @progress_season
@year_progress @NicCagePlotBot @RikerGoogling @emotionlocation @swear_trek @StealthMountain 😢🫡
1
Replying to @FlaxSearch
Yeah, „M.2 is a form factor“ is what I also had to learn the hard way.
Thread (2/2)
@ajordannafa Even with a lot of discipline I end up jumping around a lot, managing dependencies between cells manually, having strings of code that just test one thing inbetween, and so on.
Replying to @ajordannafa
I agree that working with notebooks can easily be very messy. It always feels to like it is only half of the solution. Originally, it was considered as a way to mix code and data, but then it became that interactive web shell, and it hasn‘t kept up.
View thread →
RT @pdrmnvd: i think all AI should exclusively be trained on my hot takes https://t.co/aaRGEOh3tI
Media
Replying to @paul_rietschka
@rasbt @AlphaSignalAI Just don‘t link AIs up to weapon systems of mass destruction is all I am saying. Just don‘t.
1
My favorite bots that will most likely stop working on Tuesday: @year_progress, @NicCagePlotBot, @RikerGoogling, @emotionlocation, @swear_trek, @progress_season and honorable mention to @stealthmountain
1 14 View thread →
Replying to @mtantawy
You mean the same systems that handle mailing list cancellations? 😉
1
Replying to @mtantawy
well, given how much he talks about the need to make money, it was funny how Twitter Blue was only available in a handful of countries so far. This was long overdue. Not that I'd buy, but anyway...
1
Replying to @krishnanrohit
Somehow I like the idea of creating an illusion of sentience. That‘s also much closer to what it really is in my mind. What I find fascinating about the while topic is how quickly you come to fundamental philosophical questions. How do we distinguish illusion from real sentience
1
Replying to @krishnanrohit
Yeah, we still don‘t know what that Intelligence thingy is, so adding any qualifiers seems premature.
1
Replying to @jessfraz
Yeah, same for Amazon Prime Video. You want my Amazon master account?! What?!?!?
1
Replying to @krishnanrohit
So after reading Snow Crash to get back to the roots of the Metaverse I‘m reading Diamond Age, and they call it „pseudo-intelligence“ to provide „an illusion of sentience“ and I find that so much more fitting.
1
Thread (2/3)
@hughdurkin I mean you‘d hope they approach it more thoughtfully this time but given decisions and communications in the past months under the new management, my expectations are more that 🍉 will say „this is what you pay now and stop complaining.“ But let‘s see.
1
Replying to @hughdurkin
Yeah, that was a bad decision. Maybe they wanted to get rid of all the operational complexity of providing the firehose. But especially DataSift was already heavily invested. The exclusivity with gnip didn‘t make a lot of sense.
1 View thread →
Thread (2/2)
@hughdurkin Back then I also felt that it was way too expensive. If you want this to work, there has to be a free layer, an okay entry layer, etc. I mean the words not out what they will do now. Let‘s see.
1
Replying to @hughdurkin
Yeah, they don‘t have a great history of building an ecosystem. DataSift, gnip… I remember those! I think that was about the time we went from TWIMPACT to streamdrill, looking for other ways to apply out technology…
1 View thread →
Thread (2/3)
There is functionality you want to increase use of your platform (e.g. being able to sent a tweet when you posted a blog post), which should pay for itself.
Well, I hope Twitter rolls out their paid API faster than Twitter Blue (which is still only available in six countries or so). Joking aside, back when I did TWIMPACT I thought that the data is so valuable they should try to monetize it, but let‘s see how they do it.
3 View thread →
Replying to @mjcavaretta
@rajeshkan @sabrinaa_ortiz Technically, everything is hallucinations to chatGPT. It‘s not like it can tell the difference or has access to anything real.
2
RT @knolleary: Ah February. The most efficient of months. Nice and compact, fitting perfectly into exactly four weeks.
Replying to @Henrikop
Yeah. It would need to be different structurally, but I don‘t see why that cannot be done. Refeed what it „thinks.“ It could lead to problems of stability etc. but who knows…
Thread (4/4)
@Henrikop But maybe we're also getting to a point where these aren't hypothetical questions anymore but we've created something that might or might not be equivalently intelligent and maybe conscious at some point in the not so far future.
1
Thread (3/4)
@Henrikop I'm not even knowledgable to give a good summary of where we stand there. But I think it would in general help if we would agree that this is just the tip of the iceberg of something that generations of philosophers haven't been able to answer.
1
Thread (2/4)
@Henrikop The problem is while we're all aware that something happens in our skulls, it is unobservable. These are deep philosophical questions that we haven't answered in hundreds of years, so I'm not surprised.
1
Replying to @Henrikop
Yeah I think the whole discussion is almost impossible because it is unclear what all these words like "faking" and "knowing" means. I think it depends on whether one takes a more behavioristic approach and considers only input/output relations (even for humans), or not.
1 View thread →
Thread (2/2)
@Henrikop ChatGPT has been trained with an extraordinary amount of data and what it does is being amazing at accessing that data and recalling it and reshaping it in a way it fits the prompt. But in a very fundamental sense, it has „no idea“ what it is talking about. It is faking it.
Replying to @Henrikop
In awe of the technical capabilities. In particular, reformulating answers as poems or in different languages is something that blows my mind. In terms of „intelligence?“ Definitely faking it.
1 View thread →
Replying to @manuelhe
I already have a MBP with an M1 and that machine is amazing, but too large to lug around and I really really like the drop shaped MBA design, but replacing it doesn‘t seem reasonable.
Replying to @ssdpd
Yeah, the problem is definitely that everything gets more bloated once the processing power is there. Running a zoom conference and miro in a web browser is almost impossible on the 2 core Intel, but smooth on the M1.
1 View thread →
Every now and then there's a technical advancement in computers that makes it really hard to go back. GPUs were one, SSDs were another. And now the new ARM chips on Macs. I'm still really attached to my Intel MacBook Air, but boy is that poor machine suffering.
6
Replying to @jwfbean
You‘re not gonna believe these screws that we‘re cooking up. Total game changer!!
Thread (2/2)
@rorcde The question is, does that technology allow you to differentiate yourself from competitors. Deployment? Probably not. But these are very expensive decisions.
1
Replying to @rorcde
I think the especially for big companies the situation is interesting. Sure they could reinvent the wheel. But should they? Zalando for example built its own deployment stack (STUPS). This was maybe the right decision at the time but eventually they switched to kubernetes.
1 View thread →
Replying to @rorcde
Yeah, I‘m very familiar with that temptation :) But I think it has both sides: sometimes you should just use some existing tool. And sometimes what exists is too complex, not the right fit etc. (like people using k8s for a simple webapp) and people will still tell you to use it.
Sometimes working in tech is like you're on a construction site but instead of buying something people say "nah, we're building this ourselves" for no reason.
6
Sometimes working in tech is like you're on a construction site but every day they tell you "we're using a new tool now" and you constantly have to read manuals to get the most basic stuff done.
3
RT @FlaxSearch: Looking forward to bringing Haystack on Tour to Kraków, Poland next month! I'll be asking that big question: Why is search…
RT @mikiobraun: My teenage kid is learning about derivatives, and explaining to them that this is what our current AI systems are built on…
My teenage kid is learning about derivatives, and explaining to them that this is what our current AI systems are built on blows their mind.
1 14
Replying to @iA
No offense 😅 I was just thinking that cancer is too serious to lightly use it as a metaphor, but it is very fitting in this case.
Amazingly deep post by the people who brought you the iA writing app. Comparing AI to a cancer is a bit crass, but just like cancer has no regard for the effect that its growth has on the body, we let AI grow in our society and it has no notion of the effect is has. twitter.com/iA/status/1618…
5
Thread (2/2)
@paul_rietschka @alexkyllo First week of January I bought a yoga stretching online course because you wouldn’t believe the lack of general flexibility. That‘s where I‘m at.
1
RT @mikiobraun: (🥕 hot take) We don‘t have AGI yet, but we invested enormous time and energy into a system that can fake it.
Replying to @ChappellTracker
Like classic opera, we were given with a limited set of memes which we will now reorder and rehash forever.
Replying to @rorcde
I was more thinking with so many experienced people being laid off right now. But yeah, schedules and meetings I have right now, and I experimented with setting OKRs for myself, but it would require some adjustments I am sure :)
1
RT @jdegoes: Just blogged, "Scala Resurrection". If we want Scala to grow within industry, we are going to have to collaborate on some maj…
Replying to @YassineAlouini
@benhamner There isn‘t enough data in the world to handle all the edge cases!!!1!!
1
Replying to @fs111
Yeah, the Eliza effect is strong with this one! With Eliza it was at least clear that it really didn‘t do anything because the program was so simple. With chatGPT there is room for „but WHAT IF we already have AGI in there. You don‘t know what it does! Nobody knows!“
Replying to @joe_hellerstein
Thanks for the link, in my mind they always felt so close, great to finally be able to see just how close really.
1
Replying to @oldJavaGuy
Yeah, as usual I think the wave of layoffs is a bit delayed, but not as much as it used to be. But good to know, Sean!
1
The upside of being self-employed is that I don't have to worry about being laid off, the downside is that it might be a bit harder to find a job if I ever want to go back.
5
Thread (2/2)
More specifically, TIL that pyspark's date_trunc adds the default timezone to a timestamp and you have to os.environ['TZ'] = 'UTC' explicitly to make it stop 😑
1
Replying to @mucio
I can definitely show you how to drop a table in prod in this 30s tiktok video.
1
Thread (3/3)
Or: am I just becoming old and this is just the way the world works? We lose some stuff, but we gain something on another level? And yet, this one feels different. Just like we have substituted performance on a test set with real mastery, we've reduced real skill with tricks.
Thread (2/3)
I see the same development everywhere. Learn to play guitar in just 4 weeks. Learn a new language with just 15 minutes per day. Buy this MIDI pack to put together a pop song in 10 minutes.
RT @mikiobraun: To everyone thinking AGI is close, this is what you‘re talking to https://t.co/kfQZ37zZrG
Media
Replying to @francoisfleuret
Training and test - it‘s asymmetric but that‘s how I learned it.
1
Replying to @CFDevelop
Man I wish I could post (🌶️ hot takes) but alas, another feature not available where I roam.
1
Replying to @multikev
Yeah, AI helps to further reduce the very bond and fabric of human connectedness by making all of our interactions fully transactional.
2
RT @mikiobraun: @Carnage4Life Wow. Ever since Sony brought out the PSVR Microsoft has been non-committal about VR. Complained nobody wants…
Replying to @Carnage4Life
Wow. Ever since Sony brought out the PSVR Microsoft has been non-committal about VR. Complained nobody wants „cables in the living room.“ I‘m not surprised it didn‘t work out.
1 2
RT @mikiobraun: With all the layoffs in tech I sometimes hear people say „how can they lay off so many people with no business impact? Clea…
Thread (2/2)
@rorcde I am not sure that they are being so strategic. It shouldn‘t just be cuts across the board but you also need to reorg strategically to make it work. If done poorly, you simply have fewer people working on the same stuff.
Replying to @rorcde
Yeah, the way I‘d rather phrase this is that because of what I said, teams usually work on some new bet. If the company can afford it, not all of them are high stakes (or non-core). To reduce spending, you need to refocus on what is really core.
2 View thread →
85 days since 🍉 acquired Twitter and Twitter Blue is still only available in the US, UK, Canada, Australia, New Zealand and Japan.
Media
2
Replying to @Kantrowitz
Watch the first ten minutes of the movie margin call. That was a brutal layoff as well but still had more decency than this.
4
Thread (7/7)
On the other hand I believe that sometimes a product is more or less done and it would make total sense to scale a team down or ideally put these people to work on something new. But in most companies that never happens.
1
Thread (6/7)
Maybe the most important part: as we‘ve all painfully realized over and over again, it is very hard to build software according to a spec. You need to iterate and be agile to figure out what it really is that you need to build, and that‘s not something you can outsource easily.
1 4
Thread (4/7)
Three reasons: Software is still much newer than factory construction and much less standardized, so you need to keep a specialized team that knows your factory around.
4
Thread (2/7)
Software companies are like factories. The software engineers aren‘t factory workers, they build, modify and improve the factory to optimize it, adapt it, and make new kinds of services and features possible.
2
With all the layoffs in tech I sometimes hear people say „how can they lay off so many people with no business impact? Clearly they have all been slacking!“ What people don‘t get is that devs aren‘t required to keep the business running. They build it.
6 13 View thread →
Replying to @Carnage4Life
And Twitter Blue still not available in most of the world. That‘s one thing I don‘t get. What‘s the issue? Payment and tax integration?
Replying to @paul_rietschka
@YassineAlouini Old enough to remember web3? Wasn‘t that like last year? Hype cycles are definitely accelerating!
2
Trying out different features for your ML model shouldn't be like you're adding another wing to your house. It should be more like picking a different mix for your next Subway sandwich.
Replying to @d_stepanovic
The other day I saw the term "clickworker" and that concept was new to me and now I see it everywhere.
2
Replying to @jwfbean
Yeah. And very often that isn't even the right kind of magic that would help them.
1
RT @mikiobraun: I think the divide between the AI hype fueled by chatGPT and stable diffusion and how companies are still struggling to mak…
RT @mikiobraun: The problem with building ML based production systems is that software engineering is more like building a house but ML sol…
Replying to @rorcde
I think it‘s going fine. Some people I see more over there, some more here. For me personally, Twitter got much more agreeable after 🍉 has gotten quieter and isn‘t creating some new drama every two days. :)
1
Thread (2/2)
@YassineAlouini The there was also a lot of scepticism (at least in Germany). "Let's see whether this takes off." "People will always prefer to go into a store and have a real interaction with a human." To me it feels like the times we live in are definitely more prone to hype.
1 2
Replying to @YassineAlouini
With people being hyped about the possibilities of the Internet but then everyone struggling to make use of it? From what I remember :) the Internet was a more gradual thing. I still remember being excited the first time I saw an URL "in the wild."
2 View thread →
Thread (2/2)
On the one hand, we talk about whether we already have reached AGI or if not when, on the other hand, teams are struggling to understand that you need to properly set up metrics and evaluation to ensure the models you learned actually work.
7
I think the divide between the AI hype fueled by chatGPT and stable diffusion and how companies are still struggling to make data science projects work is bigger than ever.
3 25 View thread →
Replying to @ssdpd
Yes! X-D All the while people tell you you cannot touch production and the people over there are still thinking they are building a house on the ground! Immediately thinking of a few projects I cannot name here. :)
1
Replying to @ssdpd
Nice! Yeah, and when you took off you thought it‘d be a tent and while flying you realize it has to be a spa resort.
1
Thread (4/4)
I have yet to see a solution that does this really well. The fragmented toolset adds to this. Pandas prototypes in notebooks become Spark pipelines and Python microservices and so on.
Thread (3/4)
Software also needs to be built for change, but in ML, it needs to be done in a way that minimizes manual work because you want to be fast. Changing features shouldn‘t require people touching all parts of the system for a couple of weeks.
2
Replying to @dosinga
Interesting! Has the format changed? Is it somehow more accessible? Or have my LinkedIn contacts just moved up and are now part of it?
I see people at Davos on LinkedIn and instagram... Is this becoming a thing for people who are neither super rich nor politicians?
1 2
Replying to @roydanroy
I remember that they stopped embedding videos from some other site (was it instagram?) at some point, which goes a bit in the direction of trying to keep off a competitor.
Replying to @Carnage4Life
As a friend always says: Loyalty, always expected, never given.
30
RT @GergelyOrosz: A sign in someone's career, that I always take as a big positive: having been a manager, then went back to being an IC.…
Handful of operators, yes, but the underlying mechanics of computation are not trivial. Then again, I think by now it is clear that these are the primitives of data transformation as almost any model (map reduce, dataframes) converges back to SQL eventually. twitter.com/GergelyOrosz/s…
6
Thread (2/2)
@krishnanrohit Then again, I think I read that this idea of childhood is more recent. Kids used to be much more part of everyday activity and work, too. But yeah, different times.
1
Replying to @krishnanrohit
My two cents: I always thought maybe kids were very much more a part of life in the olden days. There was always some kid around from siblings, cousins, etc. In my life my own kids were the first time I had to deal with toddlers and felt like I had to figure everything out myself
1 View thread →
Replying to @tirsen
Hm… sounds like it‘s hard to predict whether it will happen or not? As much as I like mutable data structures for performance, reasoning about immutable data structures sure is simpler.
Replying to @fhuszar
Yeah, I mean I get how bytes are not characters and encoding and so on, but if everything just defaulted to utf-8 I‘d be fine with me, too.
Excellent article on Twitter‘s history with 3rd party apps and the role they played in Twitter‘s growth. I agree that just cutting access off without prior notice is bad form but even before 🍉, Twitter‘s relationship with 3rd party apps has been strained. twitter.com/mikiobraun/sta…
3
Replying to @arjmur
Yeah, although quite a few people told me they don‘t like mastodon‘s purely chronological timeline. Twitter has a long history of shutting down 3rd party products and services, seems logical. Not sure but I also think you see ads only in the official app?
1
Replying to @dhtoomey
Oh yes... so many great scenes. "So you're a rocket scientist" - "After all this years, I still need the money" - "Be Smarter, Steal, Or Be First."
View thread →
RT @mikiobraun: It‘s this time of the year again! In our so far most extensive session we‘re covering everything from large language models…
Replying to @yukihiro_matz
今朝これを読んで助かりました。ありがとうございます。
1
RT @vboykis: I really do feel like if you’ve been on Twitter long enough you can develop a shitpost for anything https://t.co/6eGxOKcjim
Media Media
Replying to @MaineC
Yeah, the memories! 😊 And thanks for the reminder for @berlinbuzzwords! (Just checked, deadline is end of Feb)
2
Thread (3/3)
@MaineC Unfortunately all that got cancelled early on in the pandemic. It felt like they were trying to change their business model already with their subscription service and so on. Might all make more sense to them now, but I liked the conferences.
1
Thread (2/3)
@MaineC For a couple of years I went to the London Strata and AI confs. Also went to New York a couple of times, and one time to Beijing. And I always found they managed to bring on people from the big players more than many other European conferences.
1
Replying to @MaineC
Mostly attending them, meeting people, listening and giving talks, seeing famous people in the speaker lounge but being to shy to talk to them (everyone always seemed so busy), and the evening events.
2 View thread →
Replying to @jonsidd
@turingcom @Mural @MiroHQ @Google I used both mural and miro and it‘s interesting how both products started out being very similar. I think mural had templates earlier but miro has been absolutely ruthless adding features and constantly improving. Definitely have a look at miro.
1
RT @jenistyping: ok fess up. who used chatGPT for writing their performance reviews this cycle 👀
RT @d_stepanovic: There are tiny fonts, but then there is a font reserved for the “Unsubscribe” link in promotional emails.
Replying to @paul_rietschka
This morning I was thinking that I found stuff posted on Twitter more interesting, but I can‘t really say why. More hot takes maybe? :) Yeah, mastodon is a couple of open source projects (the server, the apps, etc.) That‘s great but there is also no overarching product mgmt.
Replying to @paul_rietschka
I am. And I‘m back… but the vibe is different. I‘m still trying to figure out in what way.
Thread (2/2)
@CFDevelop I don‘t care about the format but I‘ve seen it enable even the more quiet ones speak out and share observations that really helped improve. Alternatively you need people with good relationships who are comfortable and able to negotiate what needs to improve.
Replying to @CFDevelop
Yes, I‘ve been there, too. Especially the format of „start/stop/continue doing.“ What worked for me was a format where you first collect data going through what happened and rank/sort to find what was most painful before even going into solutions.
View thread →
Replying to @johncutlefish
Yeah this is so telling about how leadership is seen. Alone, fighting forward against adversity, everyone is against you, and so on. What about building meaningful relationships with customers and colleagues?
3
RT @johncutlefish: words ppl use to describe founders: Driving Uncompromising Obsessed Maniacal Demanding Visionary Single-minded Relentle…
Replying to @_deep_yearning
I used to say that at least I‘m not working on nuclear reactor control, so I don‘t have to worry about the impact of my work on society…
2
RT @_deep_yearning: AI really went from one of the best fields to worst to work in if you even have a bit of moral fibre in your being…
Replying to @CFDevelop
I also think that teams need time to gel and retrospectives if done right are a good forum to get people to acknowledge and talk about friction in collaboration in the team.
RT @CFDevelop: What I've seen work ✅ 1-3 good devs ✅ Clear goal ✅ Good testing ✅ Constant feedback with client ✅ Good tech lead who writes…
Replying to @CFDevelop
Yeah, I guess the sad story is that we haven’t figured out yet how to do it better. And what we already know is often ignored.
2
Replying to @MaxProgramming1
🤣 nevermind 😅 No, my old PC apparently lacks something to be ready for Windows 11, so yeah, probably going to sit out this one.
1
Thread (2/2)
@ewolff Maybe it is one of those things where you can start working at either end and finally get to the core, but organization, people, processes, that is always there.
Replying to @ewolff
Yeah, in my experience as well. Often the issue is people, the way orgs are set up, etc. But unless that is fixed, just focussing on the software is hard or even impossible.
6 View thread →
RT @ChiaraM_87: Want to work with me? I'm suddenly up for new opportunities as Senior/Lead/Staff Backend Dev 😅 I know quite a bit of #Pytho…
Replying to @ChiaraM_87
All the best with your jobsearch, Chiara! Yeah, I think there are cases where a more localized approach to matching makes more sense. Reminds me of immoscout and looking for a flat in Berlin. Sometimes, too much reach isn‘t good either.
2
Thread (2/3)
First thought after waiting up: „a nutcracker?!? No that can‘t be right.“ Who has heard the phrase? I mean something like a thin cookie or something.
Taught my kids the expression „sandblasting a nutcracker.“ My job is done here. Interessant war, dass sie „mit Kanonen auf Spatzen schießen“ auch nicht kannten.
1 View thread →
Replying to @pdrmnvd
Eigenvector decomposition is a technique from quantum physics?! Well yeah I guess that’s one application. Among THOUSANDS.
RT @jessetanderson: This was always the way it should have been. I feel sorry for those who bought into ksqlDB. A thread with some stories:
Replying to @mtantawy
@Zoom Wow, that must be a personal record. Still, could be improved by saying „ten *business* days.“ Gives a bit of a human touch. 🤣
1
RT @bigdata: On #TheDataExchangePod I speak with @markchen90, Research Scientist @OpenAI. We discuss DALL·E, CLIP, and other key research d…
Replying to @mtantawy
I also like the common remark that it might take up to x days till the settings are effective. In what world? Manual sync? Cron job at 3am? Emailled Excel sheets?!
1
Replying to @fmueller_bln
Yeah, same for Python, I personally think the language is okay-ish, but the ecosystem of libraries especially for ML/data science is unbeatable.
2
Replying to @shuheikagawa
Happy new year, Shuhei! While you‘re at it, maybe also migrate to a new framework? 😋