Hacker News new | ask | show | jobs
by swordsmith 239 days ago
Seems like he thinks RLVR == learning from binary reward for the whole chain, completely discounting techniques to provide denser rewards like process reward supervision?