John Quan

[lovish @louvishh]

PDR has been accepted at ICML 2026! 🇰🇷

[Charles Arnal @arnal_charles]

(1/9) Experience replay can cut LLM RL training compute by up to ~40% (without hurting final accuracy—and sometimes improving it).

Paper: https://t.co/6YcAd6EBSy

[Julia Kempe @KempeLab]

1/ #1stProof. Our second installment — this time tackling Problem 3, with @scottnarmstrong and @MunosRemi
Also check out our takeaways ��� and a short “Humor from your bot” interlude — below.