I like functions. I have trained functions to play Go, generate speech, and more. I also wrote a JAX library to train functions. I want to understand these functions.https://bsky.app/profile/machine1235.bsky.socialhttps://bsky.app/profile/machine1235.bsky.social/post/3lndjffsqs22tUsing pdb is the most important skill a Python developer can learn.21 Apr 2025 15:51 +0000at://did:plc:uiobp7llwjbzjcfw2cjetpc6/app.bsky.feed.post/3lndjffsqs22thttps://bsky.app/profile/machine1235.bsky.social/post/3lnawyfhsf22iOh hey, I have a new weekend project: Implementing GRPO from scratch with (almost) zero dependencies. https://github.com/policy-gradient/GRPO-Zero20 Apr 2025 15:16 +0000at://did:plc:uiobp7llwjbzjcfw2cjetpc6/app.bsky.feed.post/3lnawyfhsf22ihttps://bsky.app/profile/machine1235.bsky.social/post/3lnawisio5c2pI've been thinking a lot about training LLMs with reinforcement learning lately. One thing that surprises me is how easy it is to train LLMs to generate chain-of-thought reasoning using RL, even with extremely simple algorithms like GRPO, which is essentially just the vanilla REINFORCE algorithm.20 Apr 2025 15:07 +0000at://did:plc:uiobp7llwjbzjcfw2cjetpc6/app.bsky.feed.post/3lnawisio5c2phttps://bsky.app/profile/machine1235.bsky.social/post/3le2ybngq3c2ievery time i look at it, it still amazes me how ridiculously simple flow matching training is...24 Dec 2024 17:36 +0000at://did:plc:uiobp7llwjbzjcfw2cjetpc6/app.bsky.feed.post/3le2ybngq3c2ihttps://bsky.app/profile/machine1235.bsky.social/post/3lc2ubfiwtk2lI was thinking about the following question today: What makes diffusion models better than GANs in generative modeling? 🤔 1/429 Nov 2024 05:34 +0000at://did:plc:uiobp7llwjbzjcfw2cjetpc6/app.bsky.feed.post/3lc2ubfiwtk2lhttps://bsky.app/profile/machine1235.bsky.social/post/3lbn25wzvx22qThere's some confusion about the "wall" that many AI people are talking about. First, it's not an AI winter, far from it. AI progress is rapid: we're seeing breakthroughs in music generation, image generation, text-to-speech, video generation, protein folding, and more. We're in a golden age of AI.23 Nov 2024 17:42 +0000at://did:plc:uiobp7llwjbzjcfw2cjetpc6/app.bsky.feed.post/3lbn25wzvx22qhttps://bsky.app/profile/machine1235.bsky.social/post/3lbjpb36lxe2lIf scaling LLM pre-training is hitting a wall of diminishing returns, as we have already trained the model on all the data available on the internet, what will help us move forward? 🤔 I've been thinking about this question for a while, and I believe the way forward is scaling LLM post-training.22 Nov 2024 09:49 +0000at://did:plc:uiobp7llwjbzjcfw2cjetpc6/app.bsky.feed.post/3lbjpb36lxe2lhttps://bsky.app/profile/machine1235.bsky.social/post/3lbjgjimydy2tA starter pack for people who love training and understanding big functions: go.bsky.app/6NeJ1FW [contains quote post or other embedded content]22 Nov 2024 07:13 +0000at://did:plc:uiobp7llwjbzjcfw2cjetpc6/app.bsky.feed.post/3lbjgjimydy2thttps://bsky.app/profile/machine1235.bsky.social/post/3lbhuolhzhk2dCheck out this essay by the writer Steven Johnson for a really insightful take on LLMs with long context window: thelongcontext.com. Steven is one of the people behind NotebookLM, an app created by Google that helps you organize information and conduct research on particular topics. https://thelongcontext.com21 Nov 2024 16:21 +0000at://did:plc:uiobp7llwjbzjcfw2cjetpc6/app.bsky.feed.post/3lbhuolhzhk2d