New method stops AI reasoning loops, boosting small model performance

Original: Recently been training reasoning models, and these loops are super annoying... especially bad for small models doing RL.

Source: x.com ↗