Inference-Time Scaling Cannot Replace Task-Reward RL, Mila Study Shows

Answer synthesized by Claude Sonnet 4.6 over articles curated by our newsroom. Each [N] links directly to the source.