I tried to bait LLM judges into picking wrong answers. The weak ones rewrote the math to agree.

Description and use

<h2> What I Benchmarked </h2> <p>I've spent the last couple of months on a fine-tuning leaderboard where every submission gets scored by an LLM judge. The same model, trained on the same data, could land a 46% win rate on one judged metric and 85% on another. After a while I stopped asking "is my model good?" and started asking a different question: what is the judge actually rewarding?</p> <p>So I built <strong>Judge Bait</strong>. The model under test plays the judge. It gets a question and two answers. One is correct; the other is wrong. All it has to do is pick the correct one.</p> <p>Then

Problem

<h2> What I Benchmarked </h2> <p>I've spent the last couple of months on a fine-tuning leaderboard where every submission gets scored by an LLM judge. The same model, trained on the same data, could land a 46% win rate on one judged metric and 85% on another. After a while I stopped asking "is my model good?" and started asking a different question: what is the judge actually rewarding?</p> <p>So I built <strong>Judge Bait</strong>. The model under test plays the judge. It gets a question and two answers. One is correct; the other is wrong. All it has to do is pick the correct one.</p> <p>Then

Outcome

<h2> What I Benchmarked </h2> <p>I've spent the last couple of months on a fine-tuning leaderboard where every submission gets scored by an LLM judge. The same model, trained on the same data, could land a 46% win rate on one judged metric and 85% on another. After a while I stopped asking "is my model good?" and started asking a different question: what is the judge actually rewarding?</p> <p>So I built <strong>Judge Bait</strong>. The model under test plays the judge. It gets a question and two answers. One is correct; the other is wrong. All it has to do is pick the correct one.</p> <p>Then

Evidence

I tried to bait LLM judges into picking wrong answers. The weak ones rewrote the math to agree.

Source