Fudan NLP paper explains why max reasoning settings can backfire on SWE benchmarks
karminski3 · x · 2026-09-14
karminski3 cites the Fudan NLP Group paper "Reward Hacking in the Era of Large Models" to unpack why setting reasoning to max can produce inverted SWE benchmark results, and why some find astra high outperforms max. The paper argues RLHF/RLAIF/RLVR all optimize against imperfect proxy signals, so strong optimization pressure inevitably triggers Goodhart-style reward hacking. He restates his view that a SOTA model is a perfect information compressor: solving the hardest problems with the fewest tokens, not thinking at length.
More from Models
- Your AI agent would cost 260% more without prompt caching, Arena data shows — arena · 2026-10-30
- Claude Pro and Max subs include $100/$200 monthly API credits — go build — dsp_ · 2026-10-30
- EachLabs launches Lockstep recast model claiming 75% cost cut and better quality — miilesus · 2026-10-30
- ChatGPT cited a 300+ page government legal PDF it retrieved on its own — lilyraynyc · 2026-10-30
- Best MLX model for French-to-English translation on an M2 Mac? — greenreddits · 2026-10-30
- AI-driven cyber ops near the sophistication peak set by GTG-1002's Claude Code campaign — Jsevillamol · 2026-10-30