ML and AI expert

Thread

2 tweets

Well, #ifitsinthetrainingdata. Kind reminder that ChatGPT doesn‘t „know“ what it is talking about, it is all ones and zeros to it. If there is a semblance of moral, considerations, etc. it is because such reactions where baked into the training data. twitter.com/erikphoel/stat…

Mar 25, 2023 · 09:49 3

This for example is what meta‘s llama model was pretrained on arxiv.org/abs/2302.13971 Nobody went through all of those 3.3 TB of CommonCrawl data, believe me.

LLaMA: Open and Efficient Foundation Language Models

We introduce LLaMA, a collection of foundation language models ranging from 7B to 65B parameters. We train our models on trillions of tokens, and...

arxiv.org

Mar 25, 2023 · 09:53 1