New Study Probes Whether Language Models Can Report Perturbations in Their Own Activations
A new paper tests language-model introspection by training a simple linear probe to see whether models can identify where their activations have been perturbed.
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