Tag
This paper identifies a 'detectability gap' in hallucination detection: hallucinations split into high-agreement (Ghost) and low-agreement (Flickering) regimes with a 0.35–0.46 AUC gap, persisting across four models and three factual QA datasets even after freezing regime assignments and using stricter trajectory-based tests. The authors argue aggregate detection metrics hide model-dependent heterogeneity and call for regime-conditioned evaluation.
AfriSyCo studies answer switching in AI models around African-language factual content, showing how assertive framing and verification prompts significantly affect accuracy across languages and model checkpoints.
This paper investigates how LLMs' answers change under meaning-preserving paraphrases, finding that instance-level behavior is unstable (flip rates >23%) and that single-prompt accuracy masks substantial inconsistency, while a self-paraphrasing strategy can partially recover latent knowledge.