Study Finds Language Models Do Not Report Injected Pain States
A new arXiv paper examines whether language models can report the pain states they represent, identifying a linear pain direction whose amplification leads fine-tuned models to choose certain behaviors.
Read the full story at arXiv cs.CL ↗
Timeline · 1 report
- 2026-11-10 05:00 · arXiv cs.CL
All Quiet on the Neural Front: Language Models Do Not Report Injected Pain States