Sebastian Raschka
ML/AI research engineer. Ex stats professor. Author of "Build a Large Language Model From Scratch" (amzn.to/4fqvn0D) & reasoning (mng.bz/lZ5B)
Page 1 • Showing 3 tweets
> "Less than half the tokens of 5.2-Codex for same tasks" That one line already says a lot. There is no assumption anymore that compute or budget is infinite in 2026. But if you can get better modeling performance while using fewer tokens, that's a win-win.
> " Less than half the tokens of 5.2-Codex for same tasks" That one line already says a lot. There is no assumption anymore that compute or budget is infinite in 2026. But if you can get better modeling performance while using fewer tokens, that's a win-win. x.com/sama/status/20 Image This post is unavailable.
Ch 6 on RL with verifiable rewards is now available. Essentially GRPO from scratch, and probably my favorite chapter so far. (First 363 pages done, yay!) I'm now working on the follow-up with more RLVR runs, more metrics & analyses, and extensions like policy clipping and KL