How bad do you think models like Qwen3.8-27B or GLM-5.3-Flash would be with H-Neurons disabled?
Heat trend
Collecting trend data
The percentage is based on available heat signal, not comment count or independent people.
A discussion explores the impact of disabling "H-Neurons" in models like Qwen3.8-27B or GLM-5.3-Flash, based on a paper proposing this method to reduce or eliminate hallucinations. The core interest lies in how such a modification would affect the models' ability to produce functional and high-quality code, specifically wondering about their performance on benchmarks like DeepSWE if these hallucination-contributing neurons were identified and deactivated.
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?