How bad do you think models like Qwen3.8-27B or GLM-5.3-Flash would be with H-Neurons disabled?

查看原文

模型发布开源代码

qwen3

热度趋势

趋势数据积累中

百分比基于当前可用热度信号，而非评论数或独立用户人数。

AI 摘要

一篇讨论探讨了禁用Qwen3.8-27B或GLM-5.3-Flash等模型中“H-神经元”的影响，该方法源于一篇旨在通过此举减少或消除幻觉的论文。讨论的核心在于，如果识别并停用这些导致幻觉的神经元，将如何影响模型生成功能性高质量代码的能力，特别是它们在DeepSWE等基准测试中的表现。

TL;DR: this paper proposes a method to fix hallucination rates to very low levels or zero by disabling neurons which contribute to hallucination.

This discovery has been out for a while now, but it hasn't been that popular, since it kind of lobotomises parts of the LLM.

I honestly don't care too much about talking to AI, but instead care about it producing working and good code. I wonder what percentage models would get on e.g. DeepSWE if we found their H-Neurons and disabled them?