Tag
A new preprint tests whether attributing the same argument to different sources (relevant expert, irrelevant expert, non-expert, or none) changes how LLMs evaluate it, finding credibility is highly source-sensitive and that irrelevant-expertise attribution can actually reduce credibility versus no source.
This paper investigates how LLMs determine whether their answers come from a supplied document or parametric knowledge by intervening on the KV cache after answer generation. It finds that source reports are causally tied to cached Q+A state rather than the document cache, with steering directions flipping nearly all reports on smaller models.
Hermes Agent v0.20.0 introduces grounded citations that link every research claim to a specific source and passage, plus a fact-checking mode that categorizes claims as confirmed, unverified, or contradicted.
This paper benchmarks 8 LLM judges for citation quality in deep-research systems, finding that cheaper models remain competitive with frontier models on source relevance and factual support, but differ in directional bias which matters for RL training.
A product manager shares hands-on testing of Minimax M3's 1M context window on a real Q3 strategic brief, noting strong source attribution up to ~200K tokens but synthesis degradation beyond that.