04:00
2026-09-10
arxiv.org
large-language-models
Do LLMs Make More Mistakes If They Do Not Believe the Input Data?
A study posted to arXiv (2609.09363v1) found only a weak context-memory conflict in large language models, with counterfactual RDF triple inputs scoring just 0.05 lower on a 1-5 faithfulness scale tha…