evidence-sufficiency

Tag

Cards List
#evidence-sufficiency

Track, Rank, Crack: Epistemic Working Memory Scales Multi-Hop Reasoning in Language Agents

arXiv cs.LG · 2026-07-15 Cached

This paper identifies 'context dilution' as a key failure mode in language agents performing multi-hop reasoning and introduces SLEUTH, an epistemic working memory that explicitly tracks confirmed facts, active hypotheses, and open questions, showing consistent gains across benchmarks.

0 favorites 0 likes
← Back to home

Submit Feedback