Learning to Reason without External Rewards — mentioned by jimmySixDOF

Mon, 02 Jun 2025 11:23:25 GMT

RL is proving to be a weird science lately : >Spurious Rewards: Rethinking Training Signals in RLVR ### *TL;DR* We show that you can do RLVR on Qwen2.5-Math models with *completely random or incorrect rewards*, and still get massive math benchmark gains. All of the following spurious rewards give 15-20+ points on MATH-500 when RLVR training Qwen2.5-Math-7B: - RLVR + format reward (reward responses with `\boxed{}`): *+16.4%* - RLVR + incorrect reward (only incorrect answers rewarded): *+24.6%* - RLVR + random reward: *+21.4%* - (as a reference) RLVR + ground-truth reward: + 28.8% How can these…

Paper and discussion history. A mention is not an endorsement.

Learning to Reason without External Rewards — HN submission

Tue, 27 May 2025 11:19:02 GMT

Learning to Reason Without External Rewards

4 points; 0 discussion replies on HN. Paper and source history. A submission is not an endorsement.

Learning to Reason without External Rewards · HackerPapers discussions

Learning to Reason without External Rewards — mentioned by jimmySixDOF

Learning to Reason without External Rewards — HN submission