Feed Ask Research Search

Feed Ask Research Search

Home Search Ask Research Alerts

Limits of RLVR · podgraph

LO

Limits of RLVR

Work · mentioned 2 times

Research paper examining whether reinforcement learning with verifiable rewards produces new capabilities beyond those of a strong base model.

No connections recorded yet.

Latest news

RoboPapers· 17 Aug 2025
Vincent Liu says the Limits of RLVR paper shows that reinforcement learning with verifiable rewards on a strong base model does not produce new superhuman behaviours, and that letting the base model retry 100 times often beats the RL-trained model.