Study Tests Whether Language Models Can Detect Perturbations in Their Own Activations
A new paper examines language-model introspection by training models to identify where their internal activations have been perturbed, probing the gap between stored information and what models report.
Read the full story at arXiv cs.CL ↗
Timeline · 1 report
- 2026-11-13 05:00 · arXiv cs.CL
A Closer Look at Language Model Introspection: Probing Information and Reporting