Mikio L. Braun

ML and AI expert

@mikiobraun

@mikiobraun Twitter Memorial

19,827 tweets · 2008–2024 · 1046 threads

2020

RT @bigdata: 1/ In this episode of #TheDataExchangepod, @mikiobraun @jennwebb & I preview our upcoming report "2021 Trends in Data and AI".…
Replying to @FcoJerSanchez
Didn‘t know about it, thanks for telling me about it, Francisco!
1
There is a whole section of business books that should really be called survivorship bias.
1 12
Replying to @stilkov
@larsr_h Dieser plötzliche Anstieg an Anfragen war auch nicht abzusehen!
1
This got me already at „let‘s add more resources to a late project.“ People will need to be onboarded and unless you have a great plan, this will add to the communication overhead and slow the project further down. #mysticalmanmonth twitter.com/shreyas/status…
1
Replying to @mucio
I mean, do you want to do business with the company that says they „have the best tech“ or those that say they are „relentlessly customer focused?“
1
RT @garybernhardt: I wish someone would do what Heroku did, just... again. Make me structure my app in a certain way, but guarantee that I…
RT @MohapatraHemant: ~8yrs ago (Dec’12) I got a job @Google. Those were still early days of cloud. I joined GCP @<150M ARR & left @~4B (exc…
Replying to @paul_rietschka
@leonpalafox I‘m often using the car analogy myself. I think the difference is that modern cars are already quite refined products with good interfaces. ML libraries are more like combustion engines plus some sketches to let the user assemble.
3 View thread →
Thread (2/2)
@leonpalafox I also always have to think just how hard it is to understand an ML algorithm‘s implementation without knowing about the math. There is just too much going on in terms of derivations and simplifications and hacks to make it fast.
3
Replying to @leonpalafox
While I am always hoping that eventually there will be a way to safely apply ML methods without having an in depth understanding of how the algorithms work, I fully agree that ML is not just another library!
2 View thread →
Random stuff one shouldn't do over the end-of-year-break: Looking at Google scholar citation stats. I never did well in terms of publishing, but considering I left academia in 2015, I didn't do bad since then 🤷‍♂️
Media
1 13
Thread (5/5)
Okay I also threw in Ken Wilber's "A Brief History of Everything for Good Measure." (20th anniversary ed with an afterword by Lana Wachowski, this is getting more interesting by the minute...)
Replying to @peteskomoroch
True for me. I never did as many bike tours and hikes in and around Berlin than this year.
1
Replying to @fx86
Well, my interpretation ;) One core concept is „wholeness,“ showing up with your full self to work, stop wearing masks and hiding behind roles. But it also means treating one another with respect, not forcing people etc.
Thread (3/5)
Interestingly, I found myself agreeing to many parts, which might explain why I‘ve been gravitating towards overarching roles that give a lot of freedom to set up collaboration, but also why I feel like there is not any one role that is a perfect fit for me.
3
Thread (2/5)
On the surfaces, it discusses ways for organizations to self-manage, but then goes further into treating organizations as living entities and rather than trying to control and steer them, giving oneself to figuring out where they might go.
1
I finished reading „Reinventing Organizations“ today and boy it has been quite a ride. This book brings discussions of levels of consciousness together with organizational design and what it means to be a decent human being in most unexpected ways.
19 View thread →
RT @mayhewsw: What are those numbers? Oh, model outputs, just haven't applied softmax yet. Ah, cool, seems logit.
RT @truemped: Shreyas is my discovery of the year. Soo much quality content and insights and so relatable. A lot to unpack in this thread…
Replying to @truemped
I agree and find it also amazing how much information he packs into Twitter threads.
I‘m often asked about whether data scientists working at companies should be writing papers, and... it is complicated. Academic pursuit of knowledge and company goals are often at conflict to some extent. twitter.com/timnitGebru/st…
5
Replying to @ogrisel
Very cool! To be honest when they announced the switch to ARM I never expected it to be actually faster than x86! And then there is still potential from the ML cores to speed up further.
Replying to @rmminusrslash
@NanaYamazaki Yeah the app is also good on case you forgot to buy a ticket before you entered the train.
Replying to @aap_twak
@albert_cloete Do they come wired as well? I‘m having those laggy mouse issues with the Magic mouse and I‘d rather not have that.
Replying to @peteskomoroch
Strata New York last year. SAP had booked Brent Spiner to sign autographs at their booth.
2
Replying to @jessetanderson
Yeah. It's a pity, @OReillyMedia ditched the conference business.
Replying to @baggerspion
@therealpadams Same here, the more I tweet the more I‘m hemorrhaging followers 😅
1
Replying to @voidmaze
Ah, that's the one that's been out a few years, did you mean the Series X? I did not😆
Now this seems excessive. I‘ve been playing this both on the Xbox One X and the Xbox Series S and yes, it didn‘t perform great on the XOX, but also not worse than any game I ever played on PC with slightly higher than advisable quality settings for the hardware at hand. twitter.com/IGN/status/133…
1
Replying to @NanaYamazaki
I would never trust me with being able to remember it. Especially when I'm in other cities. 😓
Replying to @NanaYamazaki
That's why when I moved to Berlin after being a PhD student with a card I immediately got the Umweltticket.
1
Replying to @truemped
Das hab ich zum ersten Mal von dir 2016 oder so gehört. Und immer noch...
Replying to @Henrikop
@ian_soboroff In any case, those were your three free articles for this month, but try our time restricted offer (just 1$ for 3 months!)
Replying to @Henrikop
@ian_soboroff I forgot, did you already accept our cookies?! In any case you can click here to accept all of them or open a very detailed page with individual cookie options here.
@gsingers At least the expectations that AIs just exist is definitely there already ;) Yeah, there are many models for bits and pieces, but I find it hard to come up with something you can interact with. Maybe Siri?
Replying to @data4style
@benjamingreve Yes! I mean they probably have it easier with sensors and so on, and are probably route planning and state machines, but just being able to interact with them in realtime makes it so much more tangible.
2 View thread →
Thread (2/4)
I mean what would that even be? What is a simple AI? We have bits and pieces for parts of the challenge, but we‘re still far from doing anything tangible.
My son became more aware of AI because he has been playing video games, so I reminded him I have a „doctor in AI“ and his eyes widened and he asked me to teach him how to program an AI... and I was painfully made aware of how difficult I personally consider creating AIs.
1 11 View thread →
Replying to @timur_celikel
@gabrielhl Yes, those numbers are correct in relation. It is just a bit confusing because Germany usually talks about total number of new cases in the last 7 days per 100k people, not averages.
2
This pandemic has unlocked entirely new levels of cell phone addiction for me. I just cannot put it down and I DID NOT JUST PICKED IT UP AGAIN TO TYPE THIS DID IT?!?
6 View thread →
Replying to @mtantawy
🤯 yeah, what can I say, fun with metrics. Some of them are probably even equivalent but who knows. 🤷🏻‍♂️
1
Replying to @paul_rietschka
Well, I wouldn’t say either way to measure it is right, I was more surprised I didn’t realize it earlier.
Replying to @MaineC
Yeah. I was surprised, too. And I‘d like to think I know how to deal with metrics... 😅
1
So on like day 34532 of the pandemic I learned that Germany‘s „7 day per 100k new cases numbers“ are sums, but the US media reports averages. Meaning that Berlin‘s 212 today would be a 30 in the US 🤯
15
Replying to @peteskomoroch
You also seem to have a crazy amount of technical issues at home lately... . Coincidence?!
Replying to @fmueller_bln
But maybe draw the line at kanban boards in the kitchen? #AskingForAFriend
Replying to @MaineC
@therealpadams Yeah probably a ton of race conditions and other non-determinism.
1
Replying to @baggerspion
@therealpadams @ChrisChinch @remarkablepaper Ah, good to know. And I do!
Replying to @jessetanderson
Already have those and using them. But there is something about physically standing and organizing my thoughts that feels missing sometimes.
1
Thread (10/10)
So while I would name it as one of the formative experiences of my work life, I'm split about how important it is in reality, and would never insist people address me with my title. Unless it is in an environment where it is relevant.
Thread (9/10)
Occasionally, but only if everyone in the conversation had a Dr., we would address each others by Dr. and last name for fun, because it felt so out of place.
Thread (8/10)
After I started working at Zalando, it didn't really matter. English being the work language, everybody was using first names anyway. This is not Siemens where (I assume) a Dr. has some value.
Thread (7/10)
A few more thoughts on the importance of titles: while I worked on my dissertation, it was a really important goal. After I got it and was a PostDoc, I was in an environment where everybody got it, so it didn't really matter.
Replying to @dirkriehle
Yeah, it's definitely and very clearly an attack on the person. Regarding vanity titles, I'm aware of that that exists, but no idea how prevalent.
Okay, I think I'm past the "let's get an expensive mouse." phase. Only took like 3h. I'm getting better. Now the next question is: whiteboard for the WFH setup: yay or nay?
2
Thread (6/10)
But in summary, I have huge respect for anyone who goes through the journey of getting that academic degree, and I don't care from which school, or what the rest of the family is doing for a living. Dr. Jill Biden earned it.
3
Thread (4/10)
And finally, ironically, while I have a Dr. rer. nat. (like a doctor of natural science, you know, the hard stuff), we used to say that you cannot compare that to a Dr. med. (the MD) because typically less work goes into getting those. I know, this was a bit condescending.
2
Thread (3/10)
Or that one time I went on vacation in Spain and the "Dr." was the only thing that made it into the hotel's reservation system. I was just "doctor" through the whole stay.
5
Thread (2/10)
Which interestingly lead to the Dr. showing up on my health insurance card, which in turn has led actual doctors (physicians) be confused whether I was a colleague or not :D
2
Regarding that WSJ piece that suggests Dr. Jill Biden should drop the "Dr.", the author clearly does not know how that works. In Germany, the Dr. is the only title you can actually put on your passport, and after the painful 5 years I went to, I did exactly that.
1 9 View thread →
Replying to @mfcabrera
I have the daskeyboard tkl. „Way too expensive“ might be debatable, but the 15€ cherry keyboard I had before worked just as well 🤣
Replying to @noelwelsh
Ah, so you have some first hand experience! Do you prefer thumb or finger operated styles?
Replying to @_SyedBadshah_
Predictive clicking, yes! Like aim assist, but for UIs! Some more expensive mice claim to have an ARM processor onboard...
1
So, what is a good companion to a way too expensive mechanical keyboard? I‘ve been looking at wired mice, but it seems the main customer segment for overpriced mice are gamers?
1
Replying to @clairikine
She did the work and earned the title, I don’t care which school it is from and what any other member of her family does for a living.
1
Replying to @peteskomoroch
I am constantly amazed how cheap those arduino/raspberry pi components are.
1
@CoffeyByte @QuinnyPig Yeah. I also wonder whether they will really keep those services forever. I guess most of them have very few users. I get that they are searching for the next big thing, but it also feels they just want to claim they have the most services.
2
Replying to @alexwlchan
@DRMacIver @The_Lagrangian So they have a either types of plugs at the ends?
Replying to @16kbps
@DRMacIver How the heck are those Christmas lights constructed? Is it like an extension cord with lights?
RT @kat_heller: Samy Bengio has stood by all of us in Brain Research, and it’s meant so much. Particularly for all the women who speak up.…
Hm. Looks like I now have a chance to cross the #4444 follower count AGAIN. 😅 #goodatsocialmedia
5
RT @safiyanoble: Well done and crystal clear. “We read the paper that forced Timnit Gebru out of Google. Here’s what it says” – MIT Technol…
Replying to @fhuszar
I can recommend Google Meet. Invite >4 people and you should be good! ;)
Replying to @shuheikagawa
Okay, then the question is whether it will run Google meet without melting ;) Or throttling a lot.
2
Can someone confirm that the new M1 Apples don't turn into blow dryers running Google Meet?
4
Replying to @fs111
Now that's so refreshingly honest that I'm almost wondering whether it's parody.
2
RT @chipro: I'm of the increasing belief that the main technical challenge for companies to successfully adopt ML isn’t the lack of functio…
Replying to @fmueller_bln
Yeah, the overall effect on Berlin was probably huge. And I agree, this is all stuff that needs to be explored and learned. „Industrial software production“ is not that old as a practice it seems ;)
Thread (2/2)
@fmueller_bln I think what they got right was to make the move to microservice architectures together with the cultural change. But then they also doubled the number of people at the same time and so on ;) Interesting times. What did you think of it?
Replying to @fmueller_bln
Difficult question. In retrospect I find the amount of change that was attempted impressive. Concepts like „Team Autonomy“ were not specified clearly enough leading to adjustments later. Over the years they started adding more „ownership“ and central coordination again.
View thread →
Thread (2/2)
@Quesada I read „Connected Company“ a while back which also stressed the people on the front lines. I also have „Holacracy“ lying around. That‘ll make a good trio of books ;)
1
Replying to @Quesada
Yeah, the philosophical approach is something you need to get used to. I agree with what you, and also saw how companies become more top-down and about control as they grew. I wonder whether that‘s inevitable.
1 View thread →
Thread (2/2)
Since then I‘ve seen all kinds of approaches to leadership and organization. Interesting to take a step back and try looking at it through this book‘s framework.
1
Started reading this book. I remembered when Zalando introduced #RadicalAgility, there were a lot of discussions about this book (although it wasn‘t the blueprint for the change).
Media
7 View thread →
RT @heylauragao: "Wuhan Virus", "Chinese Virus", COVID-19. Doesn't matter the name - my hometown will forever be known for that and only th…
Oh cool, it's true what they say about the #reinvent webpage. No way to filter by language... .
After Big Sur, Safari and Chrome became even more difficult to tell apart, I think because Safari's top bar has become wider as well. Now it's like Safari: tabs on the bottom, Chrome: tabs on top. #confusing #itsjustmeright
1
Replying to @vboykis
I discovered power naps for myself when my kids were little. I didn‘t wait for the perfect moment, though. My mindset was more like „if they burn down the house while I sleep at least I won‘t be tired anymore“ 😅
2
Thread (3/4)
I still remember looking at raytracing demos on the Amiga in the 90s whose performance was more like ten pixels/second for a handful of polygons. We've come a long way in terms of computational power. The Xbox doesn't even get very warm or loud.
1
Thread (2/4)
Note how all the lights are properly reflected in the street. It is also amazing how they combine this with other rendering techniques like (I assume) bump mapping and specular highlights on the asphalt texture.
Replying to @shuheikagawa
For some reason, I just like reading about memory models and stuff.
2
Thread (2/2)
@stefan_will I thought this didn‘t exist anymore but just today I came past a page related to taxes that seems be closed between 10pm and 5am 🤦🏻‍♂️
I woke up last night being sure I have missed a meeting yesterday. I was already beating up myself for my lack of diligence, but this morning I realized that the meeting is today. Now a part of my brain suggests to miss the meeting because eventual consistency. 🤔
7
Thread (2/3)
Those days! I bought a PC without OS but with a mouse and the guy in the computer shop was like WTF. Then I installed Linux from 30 3.5" disks. The computer hat 8 MEGAbytes of RAM. I would start X if I felt adventurous to let ghostscript plot the tiger head.
5
Replying to @clairikine
And yet, the cold confidence with which Winter moves into Berlin surprises me every year.
1
Replying to @peteskomoroch
Same here. I also read about interference, but to me it also seems to be load related. When I‘m in a Google Meet it seems bad consistently. So yeah, I gave up, went back to a wired mouse...
1
Replying to @peteskomoroch
@Apple Yeah, Apple‘s own mouse is sometimes very laggy. Now I‘m using some 7€ wired mouse and it works just fine.
1
Replying to @truemped
@therealpadams But seriously, if a deadline shifts and you have to manually move all those dependencies, at least you feel the impact of that.
Replying to @baggerspion
@therealpadams It‘s what sets the 10x project manager apart from the flock.
Replying to @bobbruno70
I have a dream of copying all my hundreds of CDs to some cloud storage and writing a bespoke music streaming service for myself.
Thread (3/3)
Oh, and 3/2... I also find it interesting how every company's culture and approach to building products is reflected in the product. Amazon - relentlessly efficient but never made a personal connection. Apple - really polished, maybe a bit too much. Spotify - chaotic good!
7
Thread (2/3)
Now Spotify, the content is a bit overwhelming, but what is the single most useful feature is Spotify connect, switch between computer / phone, remote control, take your current play with you wherever you go - awesome!
3
I went through three music services in the last year. Started with Amazon Music - somehow their recommendations never did it for me. Apple Music - really liked the hand-curated feel of the lists, the slow rhythm of updates, discovered a lot of new music. 1/2
View thread →
Replying to @shuheikagawa
@kandelvijaya Aha, I think I know why you returned the work monitor! 🤯
1
RT @wilks_isaac: Oppenheimer quoted the Upanishads after the detonation of the first atom bomb, meanwhile whoever on here invents AGI is go…
Replying to @Major_Grooves
Becoming bearded and becoming German, causation or correlation?
1
Replying to @shuheikagawa
@kandelvijaya I just learned the M1 chip supports only one external monitor, though 😱
1
Replying to @JasStanford
@bourbonface Season 3 is pretty awesome. Also an unexpected reminder of our common humanity in these times.
1
Replying to @fs111
@Twitter Of course not, Twitter stopped having feature complete open APIs a long time ago... .
Thread (2/2)
It‘s funny how they just copied the instagram version almost pixel for pixel. At least they could‘ve made it more based on something like, I dunno, 140 erm 280 characters of text?
Replying to @dbentley
@slyphon I'm also realizing that US company have been doing this since way before 2020.
1
Replying to @slyphon
@dbentley Yeah, somehow I'm beginning to internalize east coast vs. central Europe, but west coast always feels so far away!
1
Replying to @baggerspion
@therealpadams @truemped I agree. Ideally, „strong opinions loosely held“ means that everyone is open about what they think, and there is constructive discussions to get to the bottom of things. But you don‘t get that without leadership work.
1
Replying to @baggerspion
@therealpadams @LinkedIn „Have you tried adding a negative vspace?“
2
RT @truemped: @mikiobraun I highly recommend the interview by @ShaneAParrish with @AnnieDuke on better decision making with uncertainty: ht…
Replying to @baggerspion
@therealpadams Also: what kind of templating engines uses dollar signs at both ends, AND allows spaces in template variables?! Or is it LaTeX?
Thread (6/7)
Amazon's working backwards approach is IMHO a good attempt at making sure there is a good amount of discussion. Disagree and commit may then be used to get moving if there isn't a clear path.
Thread (5/7)
I still believe that especially if there is a lot of uncertainty, it is better to be aware of that and openly deal with it than brushing over it in favor of busywork.
Thread (4/7)
The downside is that you loose a chance to actually get to the bottom of things. I've seen "analysis paralysis" being used without even trying, maybe because there was a worry people will get lost in the complexity of things.
Thread (3/7)
I think one of the reasons why people propose the SBLH approach is because they value action over "analysis paralysis". The same is also the original intent behind "disagree and commit" IMHO.
1
Replying to @deanwampler
@schrepfler @myfear I saw a German reply to this „Seeing this just now. Because work.“
5
Replying to @hisham_hm
@jradavenport I was wondering and totally forgot who had them first.
Replying to @IgorBrigadir
Uuuh, nice. I still remember having to put timings into some config file. But not the wrong ones, because those would've fried the monitor (actual CRTs). Or so I've been told.
1
Replying to @IgorBrigadir
I guess for Linux-on-the-desktop that's a good version of everything works :D
1
High time I‘m reading this book. I‘ve read different short versions and watched talks, but it is time for the real thing. #databaseinsideout
Media
1 20
Replying to @beaucronin
I heard a company that rhymes with boundtrout still has Hadoop on prem.
1
@holadiho @dasnuf Mein Tochter: „Ab Mittwoch sind wir auf rot.“ Ich so: „Ach echt? Ich dachte ihr redet da nur so drüber.“
1
Replying to @baggerspion
@therealpadams There is a page that lists perks but I can‘t find it right now.
Replying to @xamat
Better to lose them because of politics than math jokes (I‘ll never learn!)
1
@EtzionyYair Yes to be honest I couldn’t name a single feature off the top of my head that has been introduced in macOS in the past years. I mostly upgrade to have something shiny and new each year ;)
Replying to @paul_rietschka
Yeah I guess many apps will push cosmetic changes in the next days/weeks to adapt.
Replying to @paul_rietschka
I agree with mobile & desktop coherence. What I find interesting, though, is that they also made the icons more 3D again, unlike on iOS.
Accidentally typed Lockdon, and then thought "is that the place we're spending most of 2020?"
3
RT @SergeyFeldman: I'm glad this phenomenon has a name! The last time I observed this was while working on the Semantic Scholar search engi…
Kudos to Amazon for delivering the Xbox Series S on release day by 9:15... but... you know I have work to do and stuff... late afternoon would’ve been okay, too 😅😰😭
1 9
Replying to @johncutlefish
I haven’t lost hope yet. I have seen glimpses of this. I think. Enough to still hope.
2
I finally understood election day in the US is on Tuesday so you can have celebrations on Saturday! 😉
8
Replying to @slyphon
It is almost a step function between 45 and 50, so welcome to the club ;)
What is on my mind is how many cores I would need in the MacBook Pro 16" to comfortably run Google Meet.
8
@totopampin Because you knew all along what‘s right, right? I mean you test to validate, and you‘re never wrong, right?
Replying to @shuheikagawa
You just have to do those integrals in your head and you‘ll know what to do!
1
Replying to @fs111
I can confirm! I also liked how first you‘re like „wow this is hot let‘s see how it goes“ and then it‘s the most natural thing in the world.
1
Replying to @ChiaraM_87
@soblom I completely forgot about it and by the time I got there, the pot itself was so hot the plastic had melted off 😓
Replying to @soblom
Caught it luckily in the "coffee is overflowing" phase. But yeah, once it overheated so much, the handle fell off... 😓
RT @dmonett: "This is a reminder that GPT-3 does not do what it is not supposed to do, and that any interpretation of GPT-3 as the beginnin…
Replying to @baggerspion
@therealpadams Well good to know someone is thinking about you, eh?
Replying to @shuheikagawa
Yes! I realize switching meeting rooms is already some form of workout!
1
Replying to @shuheikagawa
A colleague of mine said she doesn‘t mind going to the office now and then except that everything is so far away from the desk :)
Replying to @fmueller_bln
@ewolff @MiroHQ I do, too! Does that mean I can approve my own PRs?
Thread (2/2)
"I tried emailing a friend who lived in North Carolina, but whose ISP was in Seattle. Thankfully, it failed. If the problem had had to do with the geography of the human recipient and not his mail server, I think I would have broken down in tears."
1
Replying to @aschrock
I saw the "1/63" but only the first three yesterday evening and thought that says it all. :D
1
Replying to @noelwelsh
I agree it was quite agreeable ;) But I always feel like they focus mostly on the rendering side. Once you start looking into state, everything gets a bit more unclear.
Replying to @peteskomoroch
I just realized that the favicon for the calendar has the date from today. And I wonder why it says 31. Surely there must be some deeper meaning to that.
1
Replying to @slyphon
I know that effect very well, even when it comes to pure thought stuff.
1
Thread (3/3)
@slyphon But that always reminds me when people have a too simplistic approach to explaining human intelligence that there is a whole lot more going on in our brains. And maybe muscles.
1
Thread (2/3)
@slyphon I also had that experience having played more or less over the years that my fingers still knew what to do although my mind started to forget. So weird ;)
1
Thread (2/2)
@slyphon But yeah, I love the stories and emotions that get attached to gear. I have a Korean made Squier Jazz Bass I bought in the 90s I played all through my university years. I spent so much time on that thing!
1
Replying to @slyphon
I always wished for a butterscotch blonde tele, but when I went and bought one that color was unavailable for the model I chose. Maybe another time 😭
1 View thread →
Replying to @peteskomoroch
I think offline for model refinement, A/B tests for the real-world check, but yeah I‘ve also had many discussions about the significance and problems with A/B tests. Even „A/B tests are just slowing us down.“ 😓
Thread (2/2)
@peteskomoroch I also think it is somewhat application dependent and on how the ML model is used. There are easier cases and more difficult cases.
1
Replying to @peteskomoroch
Yes, I‘ve noticed that as well. E.g. for recommendations one has to think about how to evaluate offline because one cannot just go back in time and show other results. But it pays off having some offline proxy metric. A/B tests take way too if you are still searching.
1 View thread →
Replying to @peteskomoroch
Very true, and it can be quite challenging to do it. E.g. if the ML model provides a score which is used to rank results, even estimating accuracy in production may not be straight forward. Looking forward to reading the article!
1
Replying to @slyphon
Yeah that‘s solid. I cannot get Logic down below 8ms or so without stutters and glitches. I like to translate that to distance from the amp, assuming sound travels 3m per ms.
1
Replying to @slyphon
Cool! How is the latency on that sim? I‘m always struggling a bit with software sim amps. I got the Yamaha THR desktop amp a few months back and that has turned out to be quite nice.
1
Replying to @xamat
Interesting read, thanks for sharing! I think Amazon and Apple have very interesting approaches to foster innovation. At least anecdotally, Amazon seems to favor a less consensus based approach where people can push something through even against resistance as long as it works ;)
2 View thread →
Replying to @baggerspion
@therealpadams Well it got buried. I admit I might've let it be buried. Because I'm really not interested.
Replying to @slyphon
Yeah, I loved Ruby, but for all reasons I think it is hell in big projects, the way you can hack and extend the language, pass code blocks to functions, etc. Python feels much more closed. Even with effort, you cannot really make code much more concise in Python.
1 View thread →
Replying to @slyphon
Yes, I am 😅 Hard to believe it is already more than 11 years old.
1
Replying to @slyphon
@rossfreeman That has always been my approach to buying gear, buy something slightly more advanced than where I am right now, learn how to use it, rinse and repeat! Have fun! ;)
1
Replying to @flueke
Ja, ich hab da auch schon drüber nachgedacht, also angenommen die werden nach 14 Tagen gelöscht, aber tauchen erst dann auf, wenn Kontakte positiv getestet wurden ergibt sich da ein etwas unübersichtliches Bild, wann das denn jetzt genau war alles... .
Thread (2/2)
@ayirpelle @djpardis @Twitter I swear, back in the old days when there still were 3rd party Twitter clients someone would‘ve written a layer that adds editing via ed commands by now.
1 2
Yesterday I missed that it‘s actually Saturday and today I missed that daylight saving time is over 🙈
1
Replying to @djpardis
Whenever I loose followers I try to think of it as „follower compaction.“
2
Replying to @stephenroller
You mean the individual states being responsible being the source of resistance?
Replying to @spelendeeend
@anas_ludens Yeah if you think about it, not that complicated, is it?
1
@berilsirmacek Yes, same for me. I don’t think I ever waited longer than a couple of minutes.
Replying to @ogrisel
Is that via mail, or do you give them some kind of form with your signature and your id card or something?
Replying to @ogrisel
Yeah I think there is a preliminary result Sunday night and then the final result in the newspapers the next day. They call it "vorläufiges amtliches Endergebnis" (preliminary official endresult). I wonder why we need three words for that ;)
1
Replying to @ian_soboroff
Oh yeah, I was mostly thinking about phones and tablets. My oldest Mac is a Mac Mini that is four years old. I had to replace the HDD with an SSD but other than that it is totally okay even with the current macOS.
Replying to @oldJavaGuy
Ah, didn't know there are mail-only states even! Yeah, I think I once discussed it and people were shocked that the German state wants to know where you live. There are other things like lack of Personalausweis attached to that I think.
Replying to @ian_soboroff
Same here. I like that I can get 5 years of updates out of devices. I had the 6s+, then got the 11. Good for another five years ;)
Replying to @dev_elizabeth
@Eli_coptero_ Yeah, I don't often use the term "best practice" but in this case... .
Thread (3/4)
You don't just register for elections, the state wants to know where you live ;) That's probably quite a different mindset from other places in the world.
2
Just saying: In Germany, you have to register whenever you move. Before an election you get a letter with which you can either go vote or request mail-in voting. On election day (always a Sunday), you go to usually a school somewhere near, wait for 10 minutes max and vote.
3 23 View thread →
Now this looks like a very interesting book. I‘ve been talking about this at conferences, but focussed more on ML production infrastructure. This book also talks about feature representation, data pipelines, reproducibility, and responsible AI in great depth. twitter.com/SRobTweets/sta…
4 24
@totopampin I can imagine 😂 Were they like „what kind of technological backwaters is this?!“
Thread (8/8)
So already in console mode, you needed hardware support. Nowadays technologies have converged and Japanese fonts and input methods you can just add if you want to.
Thread (4/8)
So this is a reminder that there are other countries and cultures that even today might have a different set of technologies which can be even more advanced than the European/US variants.
2
Thread (3/8)
They also had their own lines of home computers. As a kid, we would spend the summer vacations at my grandparents house and I would try to apply my C64 knowledge to the MSX line of computers. Good memories ;)
Replying to @slyphon
I forgot but I think my Strat kit also had the treble bleed option. Especially for Strats those highs are everything! I suspect they intentionally put a suboptimal config in there because of the price point.
1 View thread →
Thread (2/3)
@slyphon And I agree with what you said, it is great that this is so accessible! The whole process took an evening and the only thing I had to buy was the copper foil. Nice!
1
Replying to @slyphon
Welcome to the rabbit hole! 😁 I have a Squier strat which also got new pickups, but I was never happy. Then the 5way switch broke and I replaced the electronics with a prewired kits and suddenly that guitar sings!
1 View thread →
Thread (3/3)
@slyphon I was a bit surprised that there really wasn‘t any shielding at all, not even graphite. This is a Fender Vintera 60s Tele. Maybe they tried to be a bit of too authentic? 🤔
1
Replying to @slyphon
So I did the copper shielding tonight and the noise from my fingers on the pickguard have completely vanished and overall hum is also down by 75% or so - very good!
1 View thread →
Replying to @truemped
@dalatangi Monday: the Saturday after Sunday... Tuesday: the Saturday that feels like a week day, Wednesday: the Saturday where I lost count, Thursday: the Saturday that is almost the real Saturday? Friday: but this is Saturday, right? Saturday: Saturday!! 🎉
Replying to @evan
Me and a friend call this the Pile of Shame. I‘m sure I have some games which I haven‘t even unwrapped yet.
Replying to @nikovarga_hr
I‘ve been on the brink of buying one for the past months. Maybe it is time 🤔
1
Replying to @xamat
Aw man :( That piece reads like a eulogy. Keith Jarret‘s music has been with me my whole adult life ever since I started listening to Jazz.
Replying to @Major_Grooves
Heck, I'd be willing to pay his rent to not see that ad ever again. #joking #ofcourse
1
Thread (2/3)
I‘m sometimes thinking that stuff like this can happen because big companies are not free markets but can have enormous resources. So the usual market forces aren‘t at play and things that don‘t make any sense are perpetuated.
1
Replying to @joe__six
No big deal! Ich wusste nicht mal, dass Berlin einen neuen Flughafen gebaut hat!! 😂
Replying to @janapur
Die monatlichen Waffenkäufe sind ohnehin schon auf Millionenmiveau?!?
Replying to @mucio
It‘s not you it‘s not like the fact you have a lot to say. That‘s deep.
2
Replying to @slyphon
I read somewhere you should remove the ground wiring and only have the copper shielding as grounding, what do you think?
1
Replying to @slyphon
I'm about to do the same thing for my Fender Tele. It was always on the noisy side, but now it started to pick up static noise whenever I move my fingers along the pickguard...
1
Replying to @mleznik
@dhh @jasonfried Yeah, not the biggest fan of this style of business literature. Five Dysfunctions of a Team was a good read. With The Goal I struggled looking past the super stereotypical depiction of the wife of the main protagonist...
1
Replying to @simonmenashy
Took like a week for one of my colleagues to have the cigarette during standup instead of immediately afterwards ;)
2
I enjoy reading the Unicorn project, but can we talk about the amount of alcoholic beverages consumed afterhours?!
4
Replying to @peteskomoroch
@evanrsparks @dbeyer123 Yeah, I have the last name of the US president muted since 2018 or so, but still so mch goes through, never figured out what the exact rules are.
1
RT @alexhanna: Copy and paste but change what the bunny is holding (\_/) ( •_•) / > 🔪
Replying to @stephenaswanson
Yeah. I mean sometimes bad things happen and there's also a way to be respectful even after that but I always feel like nothing less than an "awesome journey" is unacceptable.
Thread (2/2)
I think what I want to say is that I'm okay with people being more open about things at work sometimes not panning out how you want them to and moving on. This should not automatically mean it's a bad employer or you're a bad employee.
6
@holadiho Da gibt es was wo mit Bilderkennung geschaut wird, wer drauf sitzt. Die Details kannst Du Dir vielleicht ausmalen...
Look, I get that it is bad form to talk badly about a former employer for a number of reasons, but I‘d prefer a bit more authenticity than „This was the most awesome job I ever had... don‘t know what will be next, but excited to find out.“
9 View thread →
If "companies that made it through a crisis came out stronger" isn't textbook survivorship bias then I don't know what is.
1 10
I'm always confusing @zacharylipton and @yayalexisgay's profile pics, probably because of the background color...
6
RT @ThamKhaiMeng: Incredible explosion of gold from a 1400 year old Ginkgo tree Photography by Han Fei https://t.co/ubwAttP1RY
Media Media Media Media
RT @IanColdwater: This isn't even about me. There are SO MANY women in tech who should've been verified a long time ago: @alicegoldfuss @sa…
Replying to @slyphon
Yeah I had one guy tell me European companies overemphasize „agile“ and all big companies work like this. This being a lot of upfront planning, global project management offices, etc. My fear is that he wasn‘t lying.
1 1
Replying to @samuel_wong_
Definitely and always ;) Those tools are also good if you want to start collecting some metrics across the company and so on. But as usual, technical solutions cannot replace actual relationships.
1
Replying to @samuel_wong_
Yeah, most tools have some minimum number of votes to even show aggregates, but if the team is small you can usually guess. Especially when there is free text input as well.
1
Replying to @samuel_wong_
Upward negative feedback is hard because of the power gradient. I think leaders need to compensate by actively inviting, building trust, etc.
1
Replying to @truemped
I think XMPP at some point was the basis for Google chat, at least I remember using some non-Google client for that.
RT @wonky_donky: woah. a library of terrible, fundamentally flawed ideas on artificial intelligence... some of these are so basically dumb…
Replying to @DJCordhose
Good question. Also, can we still trust we talk to the person we're seeing? Maybe we'll figure out new ways to "authenticate". Concerning artifacts, usual artifacts are actually pretty weird, too. I'd expect we'd get used to them and learn to filter them out.
1
Replying to @mucio
To reduce lag they would also need to be able to predict when you want to say something and what. So essentially they need to learn to read our minds 😰
Replying to @peteskomoroch
@nvidia Totally! The hype implied „replaced with AI“ though... I think this is a bit like replacing videos with a more hires version of memojis. Once you only transmit the parameters of a face, it obviously becomes a lot less data to transmit.
Replying to @paul_rietschka
Yeah, I also actually wanted to use it for recording, and the additional latency from bluetooth is a no go.
Replying to @paul_rietschka
Yes! I mean not that I have a lot of comparison, but it sounds really pretty balanced, they look pretty cool, too. And I honestly heard stuff I never noticed before.
View thread →
Investing in a pair of entry to mid-level Audio-Technica montioring head phones was totally worth it. Sometimes it's just a hint of 50Hz buzz that's left in the mix that I never noticed before.
1 5
Replying to @purbon
@holadiho Aber ja, immerhin relativ hoher Prozentsatz Maskenträger in Geschäften, aber viele Restaurants haben Tische sehr eng, höchstens die Hälfte macht Kontaktverfolgung.
1
@holadiho Ganz ehrlich, es wundert mich nicht, dass die Zahlen hoch gehen. Ich hab mich eher gefragt, wieso noch nicht.
2
Replying to @slyphon
Yeah, I recently bought a MiM Telecaster and it is 🔥. Two tone sunburst, put a black pickguard on it. So nice!
1
High point this week: got a nod of approval for my experimental AV setup with headphone and lapel mic.
4
Replying to @slyphon
Too late to turn back! But isn‘t it great how endlessly collectable and configurable guitars are while still being kinda affordable? (Depends on rate of acquisition, of course)
1
Replying to @brianloveswords
Nothing like bringing those github stats up like some piece of infrastructure being connected solely via commit hooks! 🎉
1
Replying to @tenderlove
At this point? I don‘t know anymore really. Maybe put down the phone, take a step outside, look at the trees and take a walk.
1 1
Reading a lot of English news stories these days and I have to say the Inverted Pyramid approach to structuring information makes it harder because it re-iterates all the facts with increasing level of detail several times.
1
Thread (2/3)
IIRC, it went like this: They claim asymptotically that you can transmit a certain bit rate over a noisy/power restricted/etc. channel. Proof: let's sample points at random and use these as code words. Law of large numbers shows this achieves the bit rate 🤯
1
Thread (2/2)
@karlhigley Hm. Isomorphic is of course not the right term, thinking about it, it says more that changing one will constrain the other and they eventually will have similar structure.
@karlhigley Yeah, the „spaces“ probably look pretty different, also making different things „easy“ and „hard“ depending on how you approach a problem.
Just delivered my girlfriend‘s absentee ballot to the local Späti and happily paid 3.70€. My contribution to democracy. 😤
10
Replying to @mucio
Yeah, when I did a database course at university, they went through relational algebra and all that and I thought "wow, someone really thought this stuff through" ;)
Replying to @mucio
I concur, I don't think it is a coincidence, Apache Spark eventually put a SQL layer on top.
Replying to @tonyveo
It does! 😚 Sitting in front of the TV, typing in short BASIC programs ;)
RT @martinkl: Over 100,000 copies of my book @intensivedata have now been sold. In this blog post I examine some of the economics of book-w…
Have I already mentioned that those PBT keycaps from my Das Keyboard feel like I'm typing on rocks and I like it?
3
Replying to @mucio
Always handy to have that RS-232 port for when you need to log in to your Macbook Pro.
Replying to @mucio
Ah yes to the MacBook's lack of Kensigntons, but luckily that didn't happen to me. But we had drawers once which you could open even when locked if you pushed them down a little.
Replying to @mucio
Hey but there were lockers, right? Or maybe we‘re not talking about the same company? 😉
Replying to @mucio
I bought a plastic box from IKEA to put my pens and stuff in. I still have it.
1
In the future, employers have to compete with the setup people have built in their homes (am I right, @shuheikagawa?)
3
RT @DickKingSmith: Studies have shown that listening to the sound of beavers enthusiastically munching on white cabbage can temporarily red…
Replying to @roydanroy
@justinesherry I can, too, if I slack of till Sunday afternoon before I even get started!
1
Replying to @tef_ebooks
Upwards flash is the most amazing thing ever. I had that for a while and it takes the best pictures of people indoors.
Anyone who wonders about Microsoft‘s Xbox naming scheme, have a look at this: ... 3.1 for Workgroups 3.1 95 98 2000 XP Vista 7 8 10
15
RT @sh_reya: every morning i wake up with more and more conviction that applied machine learning is turning into enterprise saas. i’m not s…
@Lauri_Apple This is so effed up. If this goes bad we have known already for quite a while.
1 1
Replying to @azeem
@_KarenHao That is true. Time to start thinking more about alternative approaches...
1
Replying to @_KarenHao
@azeem It used to be that only some companies had the data necessary to train, now it seems only some have the money to pay for the computation as well.
Exclusive deal around GPT-3, buying Bethesda and preorder launch for the new Xbox Series [SX] consoles all on the same day? Someone has been busy at Microsoft ;)
1 3
Replying to @zachdcarter
Impressive! And there is still some potential with the floor and sound insulation! Way to go!!
1
Replying to @neil_conway
Just joking. I was just updating VS Code and then wondered, why? It's not like it wasn't working before... .
Interesting observation on disagree & commit. I think this is very important, otherwise you end up with too much weight on the "and commit" part. And "disagreeing" doesn't just mean venting, but actually having a good discussion about it. How about "discuss & commit" instead? twitter.com/johncutlefish/…
2
Thread (4/4)
@DmitryKan I mean none of these can really help building up echo chambers and filter bubbles if people want to, but fixing some of the bad stuff that can happen on Twitter would be a first step.
Thread (3/4)
@DmitryKan I remember I once started reading a book from O'Reilly called "Building Web Reputation Systems" or something like it. I think they had some insights.
Thread (2/4)
@DmitryKan I am sure some sociologists are studying this and should be more knowledgable than I am ;) How does it work out on reddit, I think they have moderators who set and enforce the rules?
Replying to @DmitryKan
Well, for starters I think it would be good to provide some mechanisms, but then have some "social contract" around how you want to do it. If it doesn't make sense for people, they leave. No guarantee this will turn out healthy, but at least as a participant you have a say.
View thread →
Replying to @colinmadland
@Twitter I fear for the same reason, some algorithm trying to figure out what is „interesting.“
4
Replying to @DmitryKan
I‘m thinking in a real world discussion you can deal with people (tell them to shut up, go away, and there are some social guard rails). Something like this is missing on Twitter (and other social networks).
Replying to @DmitryKan
Interesting, I'll have a look! BTW I think one thing missing from Twitter is to control the discussion. You cannot moderate comments or exclude people from a discussion if you're not okay with them.
Replying to @DmitryKan
I used to believe you could be charged by number of followers, but now I am more concerned about having those pay who want their content be seen.
Replying to @DmitryKan
Definitely! The question is, would we be willing to pay for social media if it is ad free. And can you bootstrap such a company with all the „free“ offers already existing? It is already super hard to create a new company because of network effects as it is.
Replying to @DmitryKan
I think there are probably plenty of needs (including social media/connecting with friends and family) but we‘re currently lostly seeing ads based business models here.
Replying to @DmitryKan
I think there are a companies that write software for end users, like Jetbrains, or Adobe. Or what kinds of IT needs for independent users were you thinking of?
Replying to @jessetanderson
Yeah always hard to visualize what algorithms are doing. But you have a point, that the algorithm is really just optimizing averages over the whole user base could show even more that "it doesn't care."
Thread (2/2)
@jessetanderson But I liked the enactment with Vincent Kartheiser doing the different cost functions. „He really likes that extremist stuff, let‘s show him more of that!“
Replying to @jessetanderson
Not sure if I was already at that point in the documentary. But yeah, of course it is itself pretty filter bubble-y. The enacted story of the son being radicalized... yeah.
View thread →
Thread (5/6)
It is frightening that we might be unintentionally exploiting this human weakness especially on the big challenges that humanity faces so that some companies can make a lot money from selling bits of attention.
Thread (4/6)
Given what I know about physics and astronomy, I have a hard time understanding how someone can believe the world is flat, but I realize in other topics (e.g. how to build a career) I am quite easily influenced by others.
Thread (2/6)
Before in human history, that happened for people living in other countries, but now we have all this mobility (modulo Covid) and interconnectedness and we‘re drifting apart again.
1
The Social Dilemma on Netflix is definitely worth watching. One argument they bring is by creating those filter bubbles radicalization is amplified because people are no longer exposed to a diversity of views, dehumanizing the other side.
5 View thread →
RT @louisdorard: What should managers know about #MachineLearning, to lead projects effectively? I asked this question on social media and…
Replying to @sszuecs
What I find even more chilling is how everyone chuckles as if he just made a joke.
1 3
Replying to @monteiro
Occasionally Facebook recruiters contact me and promise I'll be able to work on the largest and most complex ML problems, and that's what I am afraid of.
Replying to @BRKeogh
I was out the other day when I saw someone make a case for properly comparing GCN vs RDNA2 TeraFLOPS.
Replying to @lawrennd
I find very little detail on how exactly they want to do it right now... . Didn't China oppose handing over the IP on the algorithms at some point as well? Or did I just make that up?
Replying to @azeem
@lawrennd Yes. Although I‘d assume it won‘t be trivial to recreate the finely tuned algorithms they have. But it seems ByteDance is already denying they are selling.
Replying to @lawrennd
Yeah but wasn‘t not sending data of US citizens to China the „reason“ for this all?
Replying to @shuheikagawa
Chrome tab manager is a bit like managing in which memory layer Android apps store their data.
1
Replying to @_fogfish_
That question is also somewhere in that thread. Answer: because we pay you money.
I think I have started to pronounce GPT as Geppetto in my mind and wonder which strange associations are at work here.
2 19
RT @MalikaCantor: Headed back to the Bay Area after spending a month here and all I have to say is: https://t.co/e95rOYSZIa
Media
RT @anniebombanie_: Me everytime someone tweets about starting a blog https://t.co/5ADpE9ec0j
Media
Replying to @caglarml
IIRC KL divergence encodes coding length if you built a code using one probability distribution as assumption and the other is the real one. Makes total sense that one being small and the other being large is bad, but the other way not so much.
1
RT @Bunny_Godfather: Someone put Blade Runner 2049 music to drone footage of San Francisco on 9/09/20 credit to Terry Tsai (YouTube) #BayA…
RT @nancywyuen: Life pretty much sucks so I'm going to do a thread of gorgeous Asians. Feel free to reply with images of the same so we cel…
Thread (3/3)
@dataScienceRet @Quesada 3) I wouldn‘t say they are afraid but you still need to know what you are doing and non-DS people lack the statistics intuition to make sure it works out of sample.
1
Thread (2/3)
@dataScienceRet @Quesada 2) pretrained models and data exists for some problems, but if you want to do something new, you need to „bootstrap data“, which is work and a bit of „dark science.“
1
Overheard from people going back to the office: „Seeing my colleagues was nice but EVERYTHING was so FAR AWAY from my desk!“
1 3
Replying to @_fogfish_
I think that puts you ahead of most of our planet‘s leadership right now.
I find the stories that are told within a community (like a company) a very interesting hint at the real values. For example, many stories from Amazon center around (a) Jeff Bezos‘s genius (b) how they didn‘t get it right the first time.
4 View thread →
Replying to @StephenPiment
I still amazed we in Germany usually have the final result the next morning.
Thread (2/2)
Not only are we humans prone to attributing intelligence, they edited and pasted those parts together to highlight the different „styles“ and „registers“?!
Replying to @Major_Grooves
@Strava Is this for running? I don‘t go „long“ over 10C, but I guess everyone is different...
RT @StephenPiment: We are now entering a zone in which people, including ML experts, will confuse increasingly powerful narrow AI with AGI,…
Replying to @nikovarga_hr
Yeah, the „physical progress bar“ of read pages vs. those still ahead is the best thing ever. ☺️
1
Replying to @shokun1n
Yes, not interchangeable. For example, I would never write notes into a book.
I’ve read many books on the Kindle and like the portability and note taking capability, but lately I have started to enjoy the haptic experience of reading a real book again.
6
Every once in a while you come across a book that is so full of original thinking that it changes the way you look at the world. "Debt: The First 5000 Years" was one of it for me. Today I learned that its author @davidgraeber has unexpectedly died.
4
RT @doctorow: I've been a freelancer - a contractor - for most of my life, but I've also been a salaried employee and an hourly employee an…
Thread (3/3)
@munterluggauer @vardi "Robots" can probably deal with that as well, or at least one of the big changes that machine learning brought to AI was that it dealt with noise. And maybe emphasized the "predictive power" aspect too much ;)
Thread (2/3)
@munterluggauer @vardi In "real life", knowledge is not clearly true or false but simplified models and assumptions which hopefully have some predictive power.
Replying to @munterluggauer
@vardi I think knowledge in traditional logic systems are already there, but you need to discover them by proving them. In that sense, there is no "new knowledge" that can invalidate "old knowledge", but just subjectively things we didn't know before.
View thread →
Reading other people's code sometimes feels like searching for a spatula in somebody else's kitchen. "Ah, there it is. But why the heck would you put it there?!?"
3 13 View thread →
Replying to @superglaze
To be honest I don't even know what it would mean to sell the US business of TikTok. Is it just the brand, move the infrastructure to US cloud providers, the whole codebase?
Replying to @shuheikagawa
I‘ve been following your journey on instagram and it has been quite an inspiration for me!
1
Replying to @fs111
Oh yeah, you need to be learning and master stuff. But the idea that there is a standard that is valid forever does not work long term.
Not sure who needs to hear this but especially in fast moving fields like computer science, it is way more important to be learning and care about doing the right thing than being an expert in a given technology. Don‘t let yourself be judged on your skills, we‘re all catching up.
4 19
Replying to @CFDevelop
Paying OSS developers for their time would go a long way towards making OSS more sustainable for the creators.
2
RT @CFDevelop: CTO: All the #OSS frameworks are 60% finished Options: a) Pay OSS person a few grand to finish a framework b) Hire contra…
Replying to @fs111
@Lauri_Apple You‘re not! One day I saw this and thought now what the heck is this. Then today when I hovered on it I saw for the first time for what. 😂
Yay, jblas and @Lauri_Apple's awesome-leading-and-managing will be preserved in the arctic!
Media
2 9
Replying to @shuheikagawa
My mother always says Germany is the country of paper. We do take the paper very seriously!
1
Looks like Berlin weather went from the tropic summer to rainy autumn without missing a beat.
8
Replying to @searls
Platform. Because it has two „tt“ in German. Compatibility (first i is an a?)
2
RT @hardmaru: How does a university based researcher keep feeling relevant in this fast changing and compute driven field? Some good discu…
RT @crichardson: How do you feel about agile now? @KentBeck: It’s a devastated wasteland. The life has been sucked out of it. It’s a few r…
Alright I think I‘ve sufficiently generalized this problem to make the solution interesting.
1 7
Replying to @mdreid
I still remember having a cross Atlantic video conference and spotting a clock on the wall that was six hours behind. But this is some next level the-earth-is-indeed-round stuff!
1
Thread (2/2)
@mleznik I put an ubuntu server image on it. Amazing little thing. So much more power than the machine I did my diploma thesis on (Pentium, 500MHz, 384MB of RAM if I recall correctly).
2
I wasn‘t aware Raspberry Pis and AWS Graviton CPUs have the same architecture. So this is my new dev machine ;)
Media
6
Replying to @dehora
Which kind of microservice architectures are we talking here? Distributed monolith or spaghetti services?
1 4
Replying to @noelwelsh
@jargnar I always felt computing is as much science as it is philosophy, so I think we‘re good ;)
2
In my experience, this is very true. People usually try to do their best within the spoken and unspoken rules. Setting those up through metrics, praise, and the stories that are told is where leaders can make a huge impact. twitter.com/copyconstruct/…
6
RT @GeeksAI: Are you passionate about #AI but don't know how to get started? After mentoring projects for @dataScienceRet @Quesada joined #…
Replying to @dtemme
Ah, okay, maybe then it's just the adapter is not properly wired up 😅
1
Replying to @dtemme
I think because there is actually no grounding available, so they put in these too big plastic pongs so you cannot fit it in?
RT @seanjtaylor: A+++++++ slides would read again (seriously this is the best thing you've skimmed all day)
@holadiho Ich hab immer noch nicht verarbeitet, dass man Legosteine auf Duplosteine bauen kann.
1
@Lauri_Apple Also a great example for German eating culture ;) Also: Eifersucht ist eine Leidenschaft, die mit Eifer sucht, was Leiden schafft. Wait, is that right? 🤔
Replying to @stephwen
@noootsab @kensuio @eugeneyan Yeah, end-to-end without proper understanding of ML is difficult as well.
1
Replying to @mleznik
Yes, definitely. It is still technically too hard to really be end-to-end.
Replying to @arjmur
@HSGStGallen Yeah... okay... no I think to be good at end-to-end data scientist there is a lot you need to be good in, pure will won‘t get you there, probably too much, too many technologies, etc. But that‘s something that will likely improve over time.
1
Replying to @arjmur
I think one thing is expertise, another is how responsibilities for roles are defined. If they are too narrow there are too many things which fall through the cracks.
Is it totally normal to consider buying an older Threadripper CPU because of its 64 PCIe lanes? #AskingForAFriend
RT @peteskomoroch: Really good thread linked here with many people at different companies describing their ML & Data Science org structures…
Replying to @purbon
@MaineC I think not having school is also not an option, but a more gradual approach would’ve been more controlable.
2
Replying to @MaineC
Yeah, we're bumping up possible exposure quite a lot. I'd prefer less of a "step change" but we'll see... .
1
Replying to @CMastication
@billdollins Reminds me I always wanted to do some Skyrim VR breaks!
RT @fulhack: Indexing your data sets and documenting it is another example of a problem lots of companies have but there’s no great tool fo…
Replying to @CMastication
„No right way just trade-offs“ should be put on some stone somewhere.
1
Replying to @mucio
@xSOLID That‘s very kind considering all the other things you have going on right now!
Replying to @_DECAF
@FlyingRoasters If only I had the mental presence to make it out, but the lack of coffee is... 😴 ;)
Replying to @baggerspion
@therealpadams @Scout24 There‘s a reason it is called „scout“.
3
Replying to @ibogost
When the pool house is occupied but you still have a bit of space left.
2
Replying to @arjmur
@try_except_ That and why does it take so long? In this day and age!?
1
Replying to @try_except_
Yep, it amazes me how bad that still is in 2020. What is it doing?!?
1
That one time I did interval nesting on table ids on databricks to figure out which row was responsible for that NullPointerException. 😅
1
Replying to @mleznik
Some web frameworks can also do autoreload well, but that is easier because the web services are stateless (well at least in principle, the server might cheat and cache data etc.)
Replying to @mleznik
I agree, one thing MATLAB nails is interactive work while you work on some code. The very simple execution model (just functions) allows it to just reload any functions in the path. I have yet to see this in Python/notebooks.
1
Replying to @DRMacIver
An kind of monitoring dashboard would also benefit from dynamic music. You‘ll know something‘s up when it goes dun-dun-duuun.
1
RT @seanjtaylor: Short thread on GPT-3. I haven't worked on text models in a long time, because (TBH) I find them boring. I had been igno…
I can't believe that even now I can still see the MATLAB behind the Python when doing data science-y stuff.
3
Thread (2/2)
But honestly, what I am most excited about is that the "simple" task of next word prediction has so many applications. But over all, models as huge as GPT-3 seem to fall into the same "ecologically not viable" technology like bitcoin, at least "at scale."
5
RT @TomerUllman: Stock photos of busy parents on the computer, rated for realism: 1: Mom has both hands on computer, child seems to need n…
RT @nasrinmmm: As amazing & amusing as #gpt3 is, this OpenAI original really reminded me of the point I used to harp on years ago: some of…
RT @kevinleeme: 1/ There's a fascinating story of a normal guy named Allen Hemberger who became obsessed with the Alinea cookbook. He spent…
RT @thecrimson: Breaking: The government has agreed to rescind DHS and ICE rules barring international students attending online universiti…
Replying to @noootsab
Martingale as in stochastic process? Wow can‘t remember when that concept last crossed my mind.
Replying to @MaineC
Definitely not programmers using other people‘s code ;) Mostly end users or companies I would say.
Thread (2/3)
@MaineC In other crafts, you can build small things and sell them. With code, people write small-ish projects, but usually they end up being open sourced libraries that can be used for free.
1
Replying to @MaineC
Well, not being fully serious as I am aware of the usual business models around OSS. On the other hand, I feel there should be a way to make money writing code that is neither being employed at a company or freelancing.
1 View thread →
Replying to @lukasruff
And also not great documentation. Sometimes I think it would be easier if we all could agree that you should just pay if you use it seriously. Also open source and free don‘t have to be the same thing. But yeah...
Thread (2/2)
@lukasruff Not criticizing them in any way, but premium features and training also means less incentive to make the OSS version great, which is a pity IMHO.
Replying to @lukasruff
I agree, I don‘t think many of the models are really working out. I‘m most familiar with Big Data companies around apache projects. Databricks for example makes money with cloud services, premium features and trainings.
View thread →
@thinkberg Looks like some patreon style feature, but still in roll-out. Here, for example: github.com/mre/hyperjson Leads to this page: github.com/sponsors
GitHub - mre/hyperjson: 🐍 A hyper-fast Python module for reading/writing JSON data using Rust's serde-json.
🐍 A hyper-fast Python module for reading/writing JSON data using Rust's serde-json. - mre/hyperjson
github.com
GitHub Sponsors
Invest in the open source projects you depend on with GitHub Sponsors.
github.com
1
RT @deliprao: Has anyone realized that this cake meme has forever polluted future image training datasets crawled from the internet? Maybe…
Replying to @mucio
@truemped @dehora @badcafe Alright, I'd say we have enough people to talk AND some sponsor for the first episode!
1
Replying to @PatrickKaster
Thanks for digging that up. Now, being faster than stock MKL based pytorch or tf as well as being more *cost efficient* I can believe.
View thread →
Thread (4/4)
I also really like how they did low level hack to make things work. IIRC the Amiga also had the video modes and the processor clock tightly interlocked.
Thread (3/4)
Later at university we had a whole lecture based on the book by Patterson and Hennesey which was also around RISC architecture, so that's definitely closer to me than x86... Nice to see it is making comeback ;)
2
Thread (2/4)
I personally "grew up" with Motorola 6502 and 68000 processors. Intel was always sort of the "evil empire". I did some assembler programming on Commodore computers and on the Amiga, but never on PCs.
1
RT @jamescham: Remember, ML is profoundly conservative and reflects the world as you present it to the system!
RT @JohnONolan: On April 29th, 7 years ago — I pushed publish on a Kickstarter campaign for half baked prototype called @Ghost — with some…
Apple takes care of older hardware longer than most other companies, so I wouldn’t stop buying it. But as much as I am fascinated by ARM, it‘ll be interesting to see how good Macs will be as developer machines in the future. twitter.com/ewolff/status/…
3
I cannot believe that even on recent hardware with PCIe solid state storage Windows updates can work for hours. In that time, you could've touched each single byte a million times. Is it CPU bound? A million retries? Timeouts?
4 View thread →
Thread (2/3)
@soblom @xamat Not saying it is baseless, he is author of some very groundbreaking papers. I guess it is a style issue whether you're publicly restating this at many opportunities or not.
Replying to @pacoid
@j_houg @chrisalbon Ah I was expecting more reality TV and I was really concerned.
1
Replying to @pacoid
@j_houg @chrisalbon Which was the basis for Logan‘s Run? Or a real life version of it?
1
Replying to @j_houg
@pacoid @chrisalbon Yeah that scifi trope where there are walled communities which are pretending everything is normal while the world outside struggles to keep alive, we‘re getting there.
1
Replying to @gabrielhl
What, they just ask you to do it and never come back to you after you sent it back?
1
Replying to @tonyveo
Ideally conciseness means that once you've wrapped your brain around it, you can focus on the essence of what you want to do without too much boilerplate code, or writing stuff very verbosely because the language does not provide the right abstractions (looking at you, Java!)
View thread →
Thread (2/2)
As a category of work/travel I liked how you visit a country and also see a bit of the work reality. Those electric outlets, door knobs, exit signs. It looks all familiar and yet so different.
6
Today I made the connection between Slack the tool and that Slack the book ("Slack: Getting Past Burnout, Busywork, and the Myth of Total Efficiency" by Tom DeMarco) and I was like "wtf happened..."
1
Replying to @peteskomoroch
@chrisalbon For a second I was like "dates in naming schemes for docker containers"?
1
Replying to @InkmiHq
@_KingOfCoders Yes! Especially when it comes to anything docker based, having that run natively makes a huge difference IMHO.
1
Replying to @baggerspion
@therealpadams Of course of course, sorry. Should’ve sticked with they!
1
Replying to @baggerspion
@therealpadams But he could've probably also just Airdropped you the pics. Did you discuss this?
Replying to @baggerspion
@therealpadams And don't get me started on the data protection side of things.
Replying to @InkmiHq
@_KingOfCoders Ja, und oft hab ich auch den Eindruck, die Call Center Mitarbeiter sind technisch auch gar nicht in die Lage versetzt irgendetwas zu tun, die können nur Kram in irgendwelche Webformulare aufnehmen... .
1
Thread (2/2)
@_KingOfCoders Ich hab vor Jahren mal einen kaputten DSL Anschluß gehabt. Nach Tagen hat mich dann irgend jemand direkt verbunden. Da war dann ein relativ kurz angebundener Techniker, der für ein paar Minuten verschwand und dann meinte "Port war kaputt. Ich hab Sie mal umgehängt."
1
Thread (5/5)
At least, compared to some hoops corporate life makes you jump through, I'd say that doing a Ph.D. paid off for me personally. Then again, universities had decades to shape the process.
Replying to @DJCordhose
I see. Yes! Ironically, being an expert means to constantly learn and expose yourself to new stuff, so struggling is part of what the role is about. Not sure people are aware.
1 2
Thread (4/5)
I'd say the biggest cultural issue is that you often feel like you cannot admit struggling with any of these. Luckily through most of my academic time I was in a team of great people and there were always people you could be real with.
3
Thread (3/5)
Sometimes it felt like actually developing strategies to deal with all these things is a big part of what you're learning while doing a Ph.D. Of course, there is often little support and you're more or less alone having to figure it all out, which is unnecessary suffering.
5
Replying to @fs111
Yeah, it's like we use email because everyone has one, and you can make it sort of intelligent (by having links to do stuff like unsubscribe). But email programs still treat it as if it is... well.. mail.
Now HEY looks interesting. I've been thinking for years that email has long become the ubiquitous transport layer for notifications across the Internet and should be treated as such by mail clients.
1 6
Replying to @noootsab
All this work on AI and I still need those little grey cells to get anything done.
1
Replying to @bobbruno70
@advincze @truemped Given the stack of layers of abstractions we‘re dealing with every day, you draw the line... at visual hallucinations?!
Replying to @bobbruno70
@advincze @truemped I‘ve been told you eventually don‘t see the parenthesis anymore.
Replying to @knutwalker
I often felt that needing to be explicit pays off to clear up the thinking. I remember cases where I thought I could just duck type it, but introducing traits made it much clearer what I was talking about.
Thread (2/2)
At least it has dependency management, supports multi-core, is hackable (macros), packages to binary executables, and is typed. No idea about conciseness, though.
Thread (2/2)
@NanaYamazaki In the end, I got some sticky foil you could put to the window which would capture them, and eventually they disappeared. Definitely don't want to repeat that!
Replying to @NanaYamazaki
Yeah, fruit flies are cute in a way. I had a horrible thing last year where (I think) I hadn't properly closed the lid on the trash can or something, but suddenly I had like a dozen new flies every day. They mostly sticked to the window thankfully and bothered me too much.
View thread →
Thread (2/2)
We‘ve definitely come a long way from ML being the enabling technology for websites based on user generated content to being a helper in radicalizing people.
1 4
Replying to @tef_ebooks
Yeah, I mostly care abouz getting useful completion hints from the IDE. So many bugs depend on the data you feed at runtime, no way around this.
Replying to @tef_ebooks
Which is somewhat ironic given that they are called software *engineers*... and yeah, what I often find hard to endure is „why aren‘t you using X“ based on the most superificial pattern matching of feature lists.
Replying to @thomasfuchs
Really, airborne transmission for a lung disease? How surprising! ;)
2
Replying to @InkmiHq
@_KingOfCoders I see an RTX 2080 Ti 🥰 but which CPU did you go for?
Replying to @baggerspion
@therealpadams Also, is there a coffee pad (to heat) connected to the USB?
Girlfriend: „How many devices do you have on you? I want to test something on Zoom?“ Me: „... just one? 😅“ #feelinglikeafailure
5
Replying to @MichalSofka
Infrastructure as code is definitly a step in the right direction.
1
I always thought being able to quickly iterate is especially important for exploratory data science work. Turns out, it's equally painful to have to wait minutes while debugging cloud infrastructure configuration...
1
Replying to @noootsab
@SparkNotebook So I'm not the only one who keeps coming back to ideas developed once? I feel like I'm still following paths that I first discovered while doing Twitter trend analysis...
So I'm learning you can actually develop a whole service in a notebook. But each cell becomes it's own file and you become "make" as you figure out which cells need to be rerun in which order to bring everything up to date.
1 11
Replying to @mucio
I‘ve been working at a Cherry KC1000 and that is so sweet. I guess I‘m too old to get used to those flat keyboards...
Replying to @smnbss
@arjmur @truemped Well, I was talking about our prior experiences of course.
Thread (3/3)
@truemped Of course, for more "transactional" websites, the risk is that this will have commercial impact. But then again, if there's no room for that, it is hard to do any form of innovative product development.
Thread (2/3)
@truemped I think LinkedIn got a lot of criticism when they revamped their whole website, and new JIRA also takes some getting used to, but ultimately, I agree that's the better path.
1
Replying to @baggerspion
@therealpadams Is this one of those cases where you actually don‘t need... waitaminute. I see what you did there ;)
1
Replying to @mucio
Yeah, "a lot", "a school", "a herd". Collective nouns are difficult in English.
Replying to @mucio
I thought the WFH con would be „you eat a lot of ice cream 🍦“ ;)
1
Replying to @channingwalton
That was a lot of thinking that happened in these 21 seconds.
4
Replying to @wrede
For someone whose company is built on the connection between people he seems to be quite out of touch with... people.
2
Replying to @smnbss
@arjmur @truemped True, but as you engage with people on Twitter, FB makes money from people who want to boost their messages. Hard to resist, I know... .
Replying to @smnbss
Yeah, Instagram and WhatsApp are the next logical steps, but not quite ready for that yet...
Thread (2/2)
Not that me deleting my account is of any relevance in the long run, but the way Zuckerberg has been putting forward a pretense of defending free speech has been hard to stomach for a long while already.
9
Alright, deleted my Facebook account. I mean I requested deletion. It will take up to 30 days (90 days) till that is done. Maybe. Some hashes might survive I guess.
24 View thread →
Thread (3/3)
Another 12 years back, in May 1996, I was in the middle of my second semester of studying computer science and math. That feels like a smoother ride in retrospect...
1
Replying to @dehora
Form some reason I cannot stop thinking about ARM, I don't think energy hungry technology can be the solution, even for AI. ;)
3
Replying to @smnbss
@arjmur Oh totally forgot about those knock off apps! They were really painful. Functional, but not the real thing! 😩
Replying to @smnbss
It was! I really liked the UI style! Those half visible tiles indicating you could swipe left *chef‘s kiss*
1
Thread (3/3)
@CMastication @GarthGrawburg I think I‘ll actually try and refer to certain roles as Hoodies from now on. „We need some Hoodies to fill in the details.“ „Can we get a Hoodie on this project?“ And so on...
1
Replying to @mucio
@MaineC Not saying SQL is bad, but it is also something we had since the 70s ;)
Replying to @MaineC
Yes, totally agree. In a way, that was the way to go to make it into a product.
1
Replying to @stadtlegende
Yeah, sometimes I think Apple is some of a few companies that really get product development. Which is interesting as anecdotally they do little of the usual customer testing etc.
2
Replying to @_fogfish_
Actually, I agree with this. I once worked in a building where they were constructing a new building across the street. I just thought if they built that buidling how we write software, the entrance would be wood, the walls would be from stone, some windows would just be MVPs...
Replying to @rishmishra
I had the same thing, turned out it was because my CPU was too old and didn't support the fancy new AVX instructions.
@EtzionyYair @Lauri_Apple Oh yes. And software being that intangible thing it is, it is hard. But we could try more I think. Unix had some good ideas, composable tools, do one thing well, etc.
Thread (6/6)
And if you're really unlucky, whoever dares to question whether all this complexity is not just bad design is judged for not having enough edge to learn it.
1
Thread (3/6)
In reality, a more polished and thought out product packaging and interfaces would allow many people to just use it, but now people are already invested in all the time they spent learning how to assemble it.
1
Replying to @noootsab
@lrz I‘m telling ya, if Twitter still allowed third party clients, I‘d be all for one which magically merges sed style edit follow ups into the original tweet.
1 View thread →
Replying to @noootsab
Can‘t you just crossports Golang‘s dependency management layer back to C++?
Replying to @noootsab
So I did a bit of reading over the weekend and C++11 looks good enough, but what still no dependency management?!?
Replying to @noootsab
Yes. I found C/C++ was a good way to learn about how computers work at a low level. Not sure you get that from languages that are too high level. Interestingly, a lot of software is still written in C/C++.
Replying to @_fogfish_
@chrisalbon I have been discussing rational vs irrational numbers over dinner, so there's that... .
Thread (2/3)
@Al_Grigor @jruby In retrospect it is interesting how jblas tried to solve a problem once (fast linear algebra on the JVM), but then tensorflow and friends took it to the next level by putting it on the GPU, and adding differentiation.
1 1
Working from home since Mar 11, I have to figure out when I‘ve been working longer from home that from the office at GetYourGuide.
4
@vetal_don Yeah I have a blog post somewhere in my mind how the combination of open source and big data startups lead to bad product quality and usability.
1
Thread (2/2)
I still remember how deeply impressed I was after watching the first Matrix movie. The rest of the evening I wasn't sure whether I am also living in a simulation and could just control traffic lights with my mind ;)
2
@adrian51gray Ah, I see. I moved the project to its own org a while back and forgot to update the links.
Replying to @azeem
Which brings me back to thinking Email has long become the universal notification infrastructure of the Internet and we need clients to reflect this.
1
These examples sound like a person who has no idea what is being talked about and makes up whatever could loosely fit the topic in the hope no one notices 😂 Which is probably exactly what you‘d get by training an ML model to imitate the training data 🤔 twitter.com/TomerUllman/st…
7
Replying to @noelwelsh
aaaah, JVM GC benchmarking, that's the JVM equivalent to bare metal :D
2
Replying to @chrisalbon
I haven‘t seen the plot yet and this is truly 😱😱😱. If a student brought me these I would‘ve had some rather critical remarks. Are they also aware that there is clearly some periodicity which is why it is going down, but only just right now?!
1
Replying to @fs111
I totally needed to get a snazzy shell theme as well just to not feel so... exposed... 😅
Replying to @bascule
Got in at 386DX40, did my master thesis on a PIII with 500 MHz and 384MB of RAM.
2
So I had a dream where setting up ziplines in Death Stranding and writing data pipelines became the same thing... time to stop playing close to bedtime... 😓
2
Replying to @DmitryKan
And now we have unicode characters in symbols... in some languages...
1
Replying to @advincze
Actually I like Scala, but python is inevitable in my profession. And Ruby was the best in terms of hackability. But maybe the answer is clojure ;) /cc @truemped
1 1
At this point, I think categories for evaluating programming languages for me are: - How concisely can I express things - Dependency management - How can I package it - Can I make use of multi-core - How hackable is the language - Types are nice
1 6
I think part of my problem is that building it myself is so much more fun than figuring out badly written documentation.
6
Replying to @DJCordhose
Viel. Alles über zwei führt zu leicht erhöhter Grundnervösität ;)
Thread (8/9)
Ah yes, one reason I think ARM is cool is because it is so memory efficient. I'm all for a 2 cm^2 of silicon die that can burn up 100W, but let's be honest, that is hardly sustainable.
1
Thread (5/9)
ARM generally sounds cool, I have an Raspberry Pi 4 with 4GB of RAM sitting around, and that machine is probably better than what I did my master thesis on.
1
Thread (4/9)
Another thing I find interesting for no reason at all are server grade ARM processors. If all you're doing is patching one REST API into another you don't need GFLOPS anyway.
Thread (2/9)
Some examples: AMD CPU/chipsets rabbit hole. I'd say this is good product management at least, everything has relatively easy to understand numbering schemes. Turns out that I accidentally bought a fitting pair of CPU and chipset (2700x and A320).
Thread (2/2)
@eigentourist We've built layers of abstractions so quickly, I wouldn't even call them "leaky" anymore. You have to understand everything to know what is going on.
Replying to @eigentourist
And yeah, I wish I could go back to C, but then I cannot live without GC, a build system, dependencies, generics, etc. anymore. But there was something nice about a language which was small enough to completely grasp.
Replying to @eigentourist
Not that exact path, but stuff like semicolons or not is the biggest hurdle to write syntactic code ;)
My daughter: “you can‘t just answer a question with ‘or’ with ‘yes’” Me: “Well, actually...” #dadandmathjokes
4
Thread (3/4)
Well, one of my old 1TB NAS drives has a new home and I didn‘t even had to install drivers under Windows or Linux 🙌 But I also saw that I‘m out if SATA power cables... (but there adapters) or places where to mount harddrives (but you could always duct tape, right? ;)
Replying to @MobileGeekGirl
Happy birthday! Same for me roughly three weeks ago, but I hope there‘ll be opportunities later this summer.
If I could send a message back in time for 500ms... I‘d fix the lag in video calls... #noyougofirst
2
Replying to @uhli
Actually I think most techniques for secret management (e.g. putting them into environment variables) are pretty insecure if you can manage to get into the server?
Replying to @evan
@therealpadams @fauna So technically they‘re more like viewices?! #scnr ;)
1
Replying to @uhli
I find it pretty intricate how to best keep production secrets hidden but have them conveniently available in production.
1
Thread (2/2)
@therealpadams @fauna And speaking of crowdfund. You mean like patreon, but for me ranting about random tech nomenclature problems? I wonder what the 5€ tier would unlock.
Replying to @baggerspion
@therealpadams @fauna I can chip in another €0.17 if it helps!
1
So turns out if I paid 15€ more, my cheapo motherboard would‘ve had an M.2 slot. But as it is I‘m out of SATA ports... sooo... time to reinstall that Windows... again...? 😅
1 View thread →
What cracks me up (a bit) is that you'd actually need to google for "less agile" and "safe agile" if you were looking into these frameworks. The first one is probably not what they're intended. And the second is a lie one way or the other. ;)
View thread →
Replying to @InkmiHq
@_KingOfCoders I think it is easier to paint something if the color stored directly, but of course having 16k colors looks nicer.
1
Replying to @InkmiHq
@_KingOfCoders Bytes are pixels row by row, either is a 3-3-2 RGB values (or probably another weighting), or an index into a color table?
Replying to @data4style
@benjamingreve So... you chose not to smile? Not sure whether I feel the need to get rid of it, it is just unusual to see because you normally don‘t look like that at yourself in the mirror.
Replying to @bocytko
Yeah, pickup points also make sense for the delivery side, so I think that's gonna be the future!
All this Zoom time has made me super aware of my facial expressions during a conversation, and in fact... I smile a lot. Like really a lot. #weird
8
Replying to @wrede
Thanks for sharing! Also: "As Jonathan Frakes would put it: No way. Not this time. It’s false. It’s totally made up. It’s fiction. We made it up. It’s a total fabrication. Not this time. It’s false. It’s a myth." 😂
@Lauri_Apple That there is no way to fix this on the road given the level of communication tech we have is a bit 1990s... .
Thread (2/2)
@mleznik Plus DHL customer service is absolutely atrocious. Every time I call they focus on telling me what I did wrong and they cannot do anything anyways.
1
Replying to @mleznik
Yes, even Hermes ;) Also having a network of small shops where to deliver to if you‘re not home makes more sense than the post office at least in my case.
1 View thread →
Replying to @bocytko
Well from a customer perspective it is nice if you know when the package will be there so they don‘t miss you and you have to go and stand in line for hours at the next post office.
Thread (2/2)
I remember DHL used to be the most advanced of the services, but I think a lot has changed in recent years with regards to planning and forecasting.
4
Okay I know these are unusual times, but I noticed some companies like DPD and GLS give me a 1h delivery window on the day before while DHL still claims my package will "probably be delivered today" when it clearly is not even close to the city I live in.
3 View thread →
Also, just because the WHO might have made a mistake does not mean everyone else did good.
1
My daughter causally remarked that the real impact of Archimedes principle was showing the importance of personal hygiene. And I cannot really argue with that.
2
RT @truemped: I wrote about efficient modeling of e-commerce campaigns (price and others) in Elasticsearch on @ThePracticalDev https://t.co…
Replying to @baggerspion
@therealpadams I still have the case from the computer I wrote my diploma thesis on. The last new motherboard didn‘t have support for floppy drives anymore, though... and IDE... #timesarechanging
View thread →
Replying to @baggerspion
@therealpadams RTX 2070 Super? Serious gaming err I mean deep learning rig!!
Replying to @baggerspion
@therealpadams What is the vertical resolution on the thing? 1080? And what is running in that PC box? Linux? Do you have GPU? Questions...!
You know what, once I'll go back to working in an office, I'll just keep the two monitor setup at home.
3
@karlhigley I‘m always amazed how some standards (like the ATX form factor) go back decades while others (like CPU sockets or RAM) seem to change every three years.
1
So I started playing Death Stranding and that is a very weird game to be playing in these days. Not just because of the post-apocalyptic nature. For example, the protagonist is mostly keeping to himself in a landscape devoid of humans, everyone else is holed up somewhere. 1/n
1 View thread →
Thread (2/2)
@karlhigley I benchmarked matrix multiplication with tf and interestingly, the 8 cores are also pretty fast so that the GPU was only about 10 times faster than the 8 cores (but about 100x than a single core).
@karlhigley I did a (somewhat smaller) thing and went for an 8 core AMD (16 HT), GTX 1660 Super, and 32 GB RAM. First time I fired up htop I was like „so. many. cores!!“ Amazing what a few hundred Euros can get you nowadays!
View thread →
RT @kncukier: Excellent article by the wise data-scientist @martingoodson on rules to govern epidemiological modeling - in public policy, a…
Interesting... when I opened Google calendar this morning first thought was "wow, these colors are really popping" #stayathome #springhassprung #itisallconfusing
1
Replying to @smnbss
Yeah, as good as the call quality is, uninstalling the thing for good is pretty hard, too, IIRC.
Replying to @jeremy_data
@bhive01 @CMastication That reminds me I also have a dynamic mic lying around somewhere 🤔
1
I coincidentally bought a cheap Cherry keyboard (approx. 13€) before WFH started and that's now what I type on for work and I like it. TBH the Apple magic keyboard is somehow too flat, I never comfortably find my way around it when touch typing.
Replying to @wrede
And throwing in void phrases like „thanks Jim, but let‘s hear what Jack has to say“
2
RT @mikiobraun: @MassimoMorelli Now‘s the time to throw all the AI we have at bridging the physical distance between ourselves!
Replying to @MassimoMorelli
Now‘s the time to throw all the AI we have at bridging the physical distance between ourselves!
1 1
Three ideas for video conferencing startups: (1) 2-3 people can break out, without leaving, audio gets a bit more isolated, (2) apply that face tracking tech for reducing bandwidth & add Animoji like features, (3) pick up from body language when people want to speak.
6
RT @Al_Grigor: I'm doing another giveaway of ML Bookcamp in parallel - on LinkedIn So far only 38 people expressed their interest. Chances…
Replying to @mucio
He is also my dentist, and we look nothing alike ;) But every other day I have someone who is ringing my bell and then a bit confused that it is so hard to get in to their dentist... .
Now that I'm spending more time at home, I realise there is a dentist next door whose name is almost like mine...
1
Replying to @data4style
@benjamingreve Hey, yes, of course. Sorry I didn‘t mean that nothing will change this, but rather (if the rate is the same) most of the infections will be new. I very much hope measures will show an effect soon, but as Isabel said, it will take a while to be visible.
2
Thread (11/11)
Moreover, apparently there has been more rigorous testing all along. Finally, I saw claims somewhere that the initial people infected were mostly younger. And apparently Germany has more capacity. Let's hope this will help...
2
Replying to @mucio
That almost sounds like you previously believed the purchase was pointless 🧐
Replying to @DmitryKan
@tomaspueyo Thanks. Yes, I think the China lines also illustrate that it is an underestimate.
Replying to @RVollgraf
Yes, saw that one, too. I found the plots comparing the real cases to number of tests in China very insightful, that led them to lock down with only a few hundred reported cases. They new the data was lagging and they were already much further ahead.
Thread (2/2)
@1meville So it is not just extremely frustrating to have difficulties being tested as an individual but it should also be in the interest of those making the decisions to have reliable data.
Replying to @_fogfish_
I remember that the Korean age distribution also looked much more similar to the general population age distribution.
Replying to @martingoodson
They plotted deaths per 100,000 over time? I think what I was interested in was comparing how deaths evolve compared to how far a country is into the outbreak, but I guess the latter is just impossible to know.
Replying to @_fogfish_
Definitely. I saw an analysis somewhere (tried to find it again nut couldn‘t) about age distribution in the cases, and Italy‘s corona cases are much older than day South Korea‘s.
Thread (8/11)
I should also add that this is not a proper mortality rate (where you would wait for cases to recover and then backpropagate that properly), but number of fatalities / number of new cases. Given how fast things are evolving, this is a proxy.
1
Thread (5/11)
Still, the differences are so huge, they don't make a lot of sense and (I suspect) rather point to huge biases in terms of how much testing is done and who is tested.
3
One thing that came up in discussions was how small the number of Corona related fatalities in Germany seems to be. So I plotted the number of fatalities vs. total cases in loglog, and indeed, the data seems very weird.
Media
5 12 View thread →
RT @TatianaTMac: Things I'd love to see more of in my timeline: 🐕 Floofs 💙 Compliments 🖌️ Art 🛋️ Couch selfies of you watching escapist TV…
@holadiho Nee, aber meine ohnehin schon trockene und vom Winter gebeutelte Haut findet's nicht cool.
Me: I really don‘t get why people think just because I‘m WFH I suddenly have so much time, still need to work, cook, buy groceries, look after my kids... Also me: boy I don‘t think I was ever this good at playing Night in Tunisia on my bass!
8
Replying to @mfcabrera
There‘s a small Nah und gut here in Friedenau who is doing a commendable job at restocking. But even they are running low now.
1
Replying to @markusandrezak
Unlike the guy who bought four packs yesterday. He probably also needs it, but I have QUESTIONS!! ;)
3
I guess it is good being called out in the street for carrying a pack of toilet paper, but IT IS ONLY ONE PACK AND I LEGITIMATELY NEEDED TO RESTOCK.
10
Replying to @tenderlove
I for one have to make it through till April 10 when the Final Fantasy VII Remake will be released ;)
RT @bbculp: Activities with low risk of contracting COVID-19: -Running -Petting your dog -Cycling -Playing with your dog -Rewatching Breaki…
RT @x0rg: Coworking from home with my partner day 1: opened the door while I was in the zone with noise cancellation headphones and touched…
Replying to @fs111
Yeah, suddenly you could listen to what you think you need and then you‘d rather follow some advice lists again...
1
Thread (2/2)
Now... anyone has good resources on some nice virtual backgrounds for Zoom? "Best virtual meeting backgrounds" image search on Google brings up way too little! ;)
4
Day two of working from home. I was pretty impressed with how well the transition was handled at @GetYourGuide. It is also amazing that with laptops, cloud infrastructure, and chat/etc. you can actually move a whole company to remote so quickly!
8 View thread →
Thread (2/3)
People suggest Japan has less steep growth because they don‘t test, but I think for that to be true they‘d need to test less over time, otherwise it‘d just be a different offset in log scale?
1
Replying to @StatsInTheWild
"Once the following day’s output was published, I grabbed that data, added it to the training data, and re-ran ten million times." Propagating prediction errors right into a feedback catastrophe. Nice! The whole wording as well... "the AI has probably not been aware of" 🙄
1 View thread →
Replying to @rbranson
How did they say in the old days? Never underestimate the bandwidth of a train full of CD-ROMs.
30
Replying to @adichad
@soblom Could be, depending on your diet. Instant ramen can only get you that far.
Replying to @mucio
No I meant isn‘t always like this, decisions percolate, you switch the db, you need to upgrade the driver version, which breaks the monitoring, and so on... hopefully it was worth it in the end 😅
Replying to @oldJavaGuy
@therealpadams I think you also put that very clearly there, and yet... ;)
1
Replying to @Quesada
Yeah, still using IntelliJ, haven’t had that problem for a while, but I guess computers are also getting faster?
Replying to @joegalarn
@jdgalarneau @namrataganatra @GCPcloud Not $20k 🤯 but yes I know stories of surprisingly expensiv BigQuery queries. I guess it is called BigQuery for a reason ;)
1
Replying to @wrede
Meine Tochter meinte auch schon, sie hätte gehört, die Hamster würden ausgehen!
1
Replying to @baggerspion
@therealpadams @NetflixDE @NintendoSwitch @rustlang You got yourself a Switch?! That‘s an awesome piece of tech, eh?
Replying to @chrisalbon
Upgrading my old Linux machine to something 6-core/16GB/lastgen Nvidia GPU with 8GB has some headspace lately, so thanks for fueling that fire. 🔥
1 View thread →
The Macbook Pro has been my work device for the past 5 years but only now that I am back to coding can I appreciate the fact that I have a 4 core/8 hyperthread CPU on my desk.
2 View thread →
A data science candidate once told me they wouldn’t allow him to move the data out of BigQuery, so he implemented decision trees in SQL and JavaScript UDFs. At which point I was like „sooo... tell me how you did it. And I mean everything...“
11
RT @jeffbigham: @mikiobraun @windx0303 I find it somewhat hilariously revealing that AI people don’t like a blackbox unexplained process de…
Replying to @windx0303
@jeffbigham At this point can't we just bootstrap AI by letting AI review the submissions and then iterate?
2
Replying to @kadavy
I've been thinking about the last point lately, and the staggering difference between the energy that needs to be put into creating vs consuming. Maybe it balances out when you consider how many people consume the creative output of a few people, but still.
Replying to @earino
People might wonder why I‘m already on the phone when I leave my home. It is to turn on some music!!
1
What is the term for the moment of disconnectedness when you‘re phone is still in your home‘s wifi but already to far away to actually receive data.
10
Replying to @earino
The kid with the fighting parents already got me before I got to the very end... .
1
Replying to @dhh
@christian_hudon Um, yep, that‘s we do in Germany. Top tax is about 45%, though, but I never considered my salary before taxes real anyway.
2
Replying to @aplokhotnyuk
@therealpadams Two times would‘ve been enough. Thanks for pointing me to it, don‘t have time right now to look into it, but I‘ll leave it on my list.
Thread (2/2)
@aplokhotnyuk @therealpadams The complexity didn‘t come from the json lib but from the data, that had odd edge cases and varying formats for the same thing.
Replying to @philgyford
It probably got bored and a bit frustrated at the missed opportunity at running something great.
RT @_KarenHao: I spent half a year digging into @OpenAI, the SF-based AI research lab, originally founded by @elonmusk. I started with a fe…
Replying to @Quesada
I do remember having a hard time with IntelliJ, turning most of the smart features off because the latency was so bad. That was few years back. Computers got faster or they got smarter at making sure that keystrokes register visually.
So my thoughts after being hands on again for about 1.5 months, everything takes as long but you need to be CONSTANTLY WORKING TO MAKE PROGRESS!
9
Replying to @fmueller_bln
Yeah, for Spark, and then I prefer to use it for more low-level stuff, but which is more on the data prep and gathering side.
Just the usual resetting the SMC on my MacbookPro so I can have two monitors connected to the thing.
Replying to @mucio
Complaining about the state of the industry? Fantasizing about something that‘s technically nice but doesn‘t solve the problem? ;)
So slack hogging about 100GB of virtual memory (but actually using only like 50MB) is... normal? I don't envy modern OSes.
Thread (3/3)
@mucio I think other than that you‘d want to get to something like airflow eventually where you can programmatically define dependencies, have proper tracking and monitoring etc.
Thread (2/3)
@mucio Next step up would be to extract the „good stuff“ into a library/jar to make the notebooks „simple“ again. Once you have that, you‘ll run into the challenge of painlessly reloading etc.
Replying to @mucio
I don‘t know man. So in my mind notebooks are like console and shell scripts. Good for interactive work and a bit of automation, but not easily testable or reusable.
View thread →
Thread (3/3)
@jessetanderson Also, and I think this is partly the „fault“ of nifty stuff like SparkSQL and streaming SQL in Flink/Kafka we haven’t really gotten much further than tables as a generally accepted mental model for data.
Replying to @jessetanderson
So in this case I‘m trying one of those invert-the-database-things and the sketch works by just keeping the data in memory. So it is wholly my fault ;)
View thread →
Replying to @j_c_cabrejas
I can definitely see both paths, but right now it is good to be back. ;)
Thread (7/8)
So on the one hand, there so much more stuff like Kafka, Kubernetes, and so on. On the other hand, I‘m still copying XML configs from the Internet together to get a sane logging setup.
1 6
Thread (6/8)
And I was also surprised that it feels like I‘m continuing just where I left off 4.5 years again. Maybe because I am now working again with event data, but I am again parsing JSON, keeping GB in memory and trying to make things fast.
3
Thread (5/8)
So on the one hand it is a bit scary how much time it can take to fix some bugs (especially me being a bit rusty), but then again, the overall speed and progress is quite comparable. I mean how much really gets done in a one hour meeting?
7
Thread (4/8)
Then, maybe two weeks ago I suddenly realized that I could really dive in and get some work done each day, and that I had almost forgotten how that feels like during the highly fragmented days of an Engineering Manager.
8
Thread (2/8)
@GetYourGuide After having spend the last 4.5 years at Zalando with high to very high meeting load, just sitting at a desk, coding away at a computer required a bit getting used to.
6
So one decision I made was to become more hands-on again when I joined @GetYourGuide. I know it is still early, but it has been... interesting... (not in the Firefly sense).
14 View thread →
Replying to @lemire
@IgorBrigadir @richardstartin Wow, GB/s. I cannot even imagine where JVM based JSON parsers are... .
1
Replying to @noelwelsh
Aaaaah! Do you think this will be just as quick in terms of JVM bytecode?
TIL that scala's mapValues calls the function on retrieval, every time. Which also means, any exceptions will triggered every time you use the map, not when you construct it.
1 View thread →
Replying to @ivan0yu
@_fogfish_ Now, think doing C/C++ for hardware/real-time stuff makes total sense, maybe even not having a GC. The discussion above was about languages for NN, which is more numerical computations, and I think there, you need to go deeper.
RT @Major_Grooves: I'm pretty sure Scrum, Agile and Design Thinking were all developed by 3M to prop-up their ailing Post-it note business.
Replying to @_fogfish_
I think golang is similar to C in terms of abstracting away from the bare metal.
Replying to @aplokhotnyuk
@truemped I haven't! Actually, it would be interested if this went beyond just JSON (most JSON libraries have some version of this implemented, together with casting stuff to case classes). But data could also come from avro, thrift, etc.
View thread →
I do love Scala but when dealing with dynamic data (etc. JSON stuff), it is quite painful. Maybe time for clojure after all, @truemped?
5
Thread (3/3)
@hisham_hm @justincormack @lproven @kmett Very interesting thought what it would take to map all of that into a language. But C is definitely a waaaaay too simplified view of modern CPUs to give you a handle to control this.
1
Thread (2/3)
@hisham_hm @justincormack @lproven @kmett ... while also organizing the flow of data (GBs of data) such that you exploit locality in a way that you don't get slowed down by the memory hierarchy. And then there are vectorized operations, and nowadays GPUs of course, and so on.
1
Replying to @hisham_hm
@justincormack @lproven @kmett I only went into this a bit while working on jblas. Numerical workloads like matrix-matrix-multiplications are interesting because on the one hand you can exploit the parallelism from having multiple FPUs in a modern CPU ...
View thread →
Thread (2/2)
IIRC GotoBLAS optimized for cache misses of the TLB, which is a thing that just transparently exists but has no explicit representation on the coding level.
3
RT @nickm_tor: Gaze not into the abyss, lest you become recognized as an abyss domain expert, and they expect you keep gazing into the damn…
Replying to @noootsab
I completely forgot there is a whole subgenre of Scala projects about mapping untyped data to case classes.
Replying to @noootsab
CPU burning cycles parsing JSON, VisualVM to see how memory consumption is going. This is like the good ol' days!
Replying to @lalleal
You mean: Me explaining what the order of the conversation on my Twitter thread is.
Media
2
Replying to @octonion
I think notebook somehow evolved from ideas of literate programming (a la Knuth)/unix shells/shell scripts/shells with graphics output/mainframe/computing but now it is used as the one user interface for data scientists by some. So it is a bit messy. ;)
1
Replying to @octonion
Yes, I think whole categories of how you can structure code usually are missing.
1
Replying to @rvdgeer
Thank you very much! Yes, they will be available on the conference website... if not let me know...
1
Replying to @compArch1
Unfortunately I‘ll already leaving this afternoon... and no worries ;) so many interesting talks!
Replying to @rorcde
@RodrigoRivr Just here today. They already had two days of workshops on the weekend. I think it is closer to O‘Reilly‘s AI conference. And more focus on banking and pharma. But definitely comparable!
1
Replying to @lalleal
@_fogfish_ @joelgrus IMHO one ingredient is also a lack of understanding/simplified view of what data scientists do on the side of the people building data tools and infrastructure.
At #AMLD2020, a substantial amount of people is working on tensorflow code in dark mode while attending the keynote. From a distance, syntax highlighted Python code is quite nice aesthetically.
4 View thread →
Replying to @larsgeorge
No offense taken, mathematics was only my minor topic ;) But is that a true story, I didn‘t know!
Replying to @_fogfish_
@lalleal @joelgrus Yeah, but as I said, databrickes et al have some features like scheduled notebook runs for production.
Replying to @_fogfish_
@lalleal @joelgrus I‘m often thinking about that, too. I think because that experience was so rich. It had everything...
Thread (3/3)
@_fogfish_ @lalleal @joelgrus Of course, there are ways, you can put stuff in libraries and then import them, and so on, but I think there is a steep change in complexity if you start doing that. I may be wrong, but that is how it looks to me currently.
1
Thread (2/3)
@_fogfish_ @lalleal @joelgrus So what about unit tests, what about code reuse? What about abstractions? What about collaboration. And many DS don‘t have a software engineering background, which makes matters worse in terms of production code.
1
Replying to @_fogfish_
@lalleal @joelgrus What I am thinking about is how databricks/sagemaker and others are selling notebooks as the one-env-for-everything, making them schedulable, callable, etc. Which is cool, but they are also just a bunch of shell scripts essentially. Living in a file system.
1 1 View thread →
Replying to @americanwombat
Gawd those dwarfes. I got claustrophobic just listening to them complaining about the lack of confined spaces outside.
I just came up with the term "AI mind clones". Is that what happens if you train one of those NLP models on personal communication?
Replying to @avibryant
„Independent newbie AI researcher“?! 🤣 Kudos to this man! It is so hard to go back and start with knowing nothing.
Replying to @earino
When I‘m old, I expect Misa to have grown up and take care of me so that my family doesn’t have to constantly remind me who I am.
Replying to @lalleal
@joelgrus That looks good, will watch! Unfortunately that is a discussion that comes up way too often with tool makers. It is like going back to 80s BASIC. Great for interactive tinkering, but you throw out decades of knowledge... unit testing, versioning, collaborations, etc.
2 3
Replying to @lalleal
I agree but tool makers seem to think it is the ideal dev environment.
1
Is there a style guide for how to write good Jupyter/databricks notebooks? 'Cause I have some strong opinions on the matter. ;)
1 10
Replying to @fs111
Well, it says I have LTE but it feels like that is connected to a 28.8k dial-up? OTOH I could also take the U2... it would take longer... but Internet! 🤔🧐
Replying to @ChiefScientist
I once worked with a Vlad (who was Romanian), and he was awesome!
So I don‘t mind the length of the commute but the spottiness of mobile Internet in the S-Bahn tunnel under Mitte is... very pronounced.
Replying to @vivekjuneja
@JadejaDA @koolhead17 @parolkar @abhishekvarier @hishami @timkrgr @phelo @dehora @clairejlew @therealpadams @mrgnrdrck I‘ve been thinking about company culture a lot, and what is good for the company, and good for the people. Not sure these are naturally the same.
Replying to @vivekjuneja
@JadejaDA @koolhead17 @parolkar @abhishekvarier @hishami @timkrgr @phelo @dehora @clairejlew @therealpadams @mrgnrdrck Some first thoughts: How you cannot solve communication problems with tooling? What it takes to co-create? What is the right metaphor for writing software? And a lot about culture I guess.
2
Replying to @vivekjuneja
@JadejaDA @koolhead17 @parolkar @abhishekvarier @hishami @timkrgr @phelo @dehora @clairejlew @therealpadams @mrgnrdrck Very interesting! I have to ponder...
1
I think we need a new kind of sentiment analysis for Twitter: - righteously enraged - defeatistically sarcastic - stubbornly idealistic - openly aggressive - ignorantly hyped Anything else? Also, how would you classify this tweet?
1 14
So half of my instagram stream seems to be at NAMM and I‘m thinking back to all the times my mother told me I shouldn’t go into music because I‘m not good enough.
1
Thread (2/2)
@earino To me, how you think about the problem you want to solve is the main difference between an engineering and a data project, not the amount of ML.
Thread (3/5)
But then again, isn't a Linux server like some very abstract version of a Minecraft server? People can log in and collaborate, and sometimes you have to kick someone of the server :-P
Thread (2/5)
So I told my son about the console and commands like "ls" and "cd." He said "oh I know the console from Minecraft." Tried to explain that this is the "Linux console." Let's see how this goes.
Replying to @baggerspion
@therealpadams @MeriamKharbat You‘re like... the SCRUM whisperer.
Frameworks that demand a whole life‘s dedication to master them. You know you‘re just part of a tool /chain/?!?
1 5
Replying to @Pantouffles
@elle4short @exfatalist But good news, it doesn’t get worse after you‘re 50!
1
Does anyone remember pheed.io? They wanted to create a pay per view social network and got a few celebs to sign up. Soon they were sold for $40M to mobli. I just learned that story is still ongoing... timesofisrael.com/beitar-jerusal…
pheed.io · No preview available
Beitar Jerusalem owner sued over hugely hyped, now defunct photo-sharing startup
NIS 6m claim alleges Moshe Hogeg and associates misled investors about Mobli app’s financial strength and about celebrities who’d put money in it;...
www.timesofisrael.com
View thread →
Replying to @truemped
@DKB_de Gestern noch gescherzt, dass manche Kunden sind den JavaScriptcode anschauen würden um zu schauen, ob eine Firma vertrauenswürdig ist... ;)
3
Replying to @choongng
What, and not totally mess up my Linux install!? (Just kidding, thanks, will have a look!)
Trying to get tensorflow-gpu to run on my old Linux machine which has recently gotten a cheap nvidia card. Turns out the problem is the ten year old AMD CPU - no AVX. 😂 But still good enough to run minecraft so my son can continue honing his redstone skills.
View thread →
RT @noootsab: Help me spread this word out the network:-D. We, @kensuio, are looking for a #Chief #Data #Product #Architect for our #DataA…
Replying to @rorcde
@RodrigoRivr Well, too early to tell, but of course each company is its own microcosm, so they are all different. Very obviously, a recurring topic is vacations and experiences (motto is #lovewhereyouregoing), whereas at Zalando it was more about fashion and self expression.
1
So after two days of onboarding at GetYourGuide, I can report that people are nice, humanity isn't lost, and I couldn't stop myself from cracking a joke to defuse awkward silence if my life depended on it.
14 View thread →
Replying to @peteskomoroch
@GetYourGuide @mathieubastian Oh right, the old LinkedIn connection! I will!
So after some time off, today (well technically yesterday) I start my new position as Staff Data Scientist at @GetYourGuide. 🎉
1 62