3 ms·
Does RL Incentivize Reasoning Capacity in LLMs Beyond the Base Model?
- yorwba 2y agoDiscussed two days ago: https://news.ycombinator.com/item?id=43760625 https://news.ycombinator.com/item?id=43760625
- falcor84 2y agoFunny how this title follows Betteridge's law of headlines, in this case demonstrating that RLVR (Reinforcement Learning with Verifiable Rewards) doesn't help the model generalize, but rather seems to overfit it, reducing the overall reasoning capacity.