krkartikay

3 ms·

1. 1.▲
   Avatarl: Training langauge models from scratch with pure RL(tokenbender.com)

   2 points by krkartikay 1y ago|0 comments
2. 2.▲
   by krkartikay 2y ago

   I tried writing an AlphaZero clone to play Chess on my home PC (I only had an RTX 3070) and I failed for essentially the same reason as they mentioned: iteration time was too slow and you couldn’t tell if the model was getting any better at
3. 3.▲
   by krkartikay 3y ago

   Pessimistic take: after the two stages you mentioned, in the third stage they will carry on BSing as usual (only more efficiently with LLMs now) and there will be no one to point out that the emperor is naked.
4. 4.▲
   by krkartikay 3y ago

   That is not the first rule in the book. Although yes Jordan Peterson might say something like that. Just to get the facts clear: the first rule in his book states “Stand straight with your shoulders back.” which he discusses both in its lit
5. 5.▲
   by krkartikay 3y ago

   Video summary: https://www.youtube.com/watch?v=BrjAt-wvEXI I wonder how long it is until someone finally figures out how to combine MCTS + RL (like AlphaZero) with LLMs and it's game over for humans. I truly think that
6. 6.▲
   Tree of Thoughts: Deliberate Problem Solving with Large Language Models(arxiv.org)

   34 points by krkartikay 3y ago|2 comments
7. 7.▲
   by krkartikay 4y ago

   I had seriously thought the title was a reference to GPT 4.
8. 8.▲
   Ask HN: Best hardware investment for learning ML in 2023?

   7 points by krkartikay 4y ago|2 comments