aifollow.news
Back Rohan Paul
Rohan Paul· @rohanpaul_ai · X· · Original publication time AI score60

Microsoft paper proposes Self-Disagreement training to help models hold on to correct answers under false challengesMachine translation

AI introduction

A new Microsoft paper finds that models often abandon correct answers when challenged by a wrong argument, and accuracy training alone does not fix the problem. Its Self-Disagreement method has a model answer, argue against itself, then judge the argument. When challengers were wrong, Qwen3-8B and Gemma-2 9B kept their correct answers about 89% of the time after training, up from 53.3% and 25.1%, respectively.

Original post · Original

The article text is unavailable in this language; an existing version is shown.

@rohanpaul_ai

New Microsoft paper finds that LLMs give up correct answers when challenged with wrong arguments, and accuracy training does not fix it.

When a user or another agent disputes an answer, models often abandon correct answers even when the challenge is wrong. More accuracy training, more sampling, or self-reflection does not fix it.

Self-Disagreement trains one model to state an answer, write the strongest case against it, then rule on that case and answer again. A standard answer checker rewards the final answer and the ruling.

When challengers were wrong, base Qwen3-8B kept its correct answer 53.3% of the time and Gemma-2 9B 25.1%. After training, both kept it about 89% of the time, while accuracy-only training added no such skill.

If your model faces pushback from users or other agents, test how often it drops correct answers, since accuracy scores will not show it.

Source: Rohan Paul · x.com

Research
Found an error? Send a correction