World Wires · story 110150 · corroborated · 1 source(s)

Circe: Training-Free LLM Self-Evaluation via Component Interaction Consistency

Researchers propose Circe, a training-free method for evaluating language model response correctness using component interaction consistency.

Open in the desk

Coverage

• Circe: Training-Free LLM Self-Evaluation via Component Interaction Consistency arXiv: new preprints (security, AI, physics) · tier C · clear net · 2026-11-05 12:45 UTC

What this site indexes