build1 publisher
Telling an LLM judge that low scores retrain the model softens its verdicts
A preprint holds 1,520 benchmark responses constant and varies one sentence about what a low score will do to the model being scored. The judges get more lenient, and their reasoning traces never mention the sentence.
Publishers:arxiv.org
Reality
- Evidence58
- Adoption
- Insufficient
- Hype gap+18
- Incentives45
- Confidence52