Y
Hacker News
new
|
ask
|
show
|
jobs
by
swordsmith
192 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?