Reflqt.
HOMERESEARCHPUBLICATIONSBLOGEXPLOREABOUTCONTACT →
§ art-012-multi-hop-reasoning · April 18, 2026 · LITERATURE-REVIEW

Multi-Hop Reasoning in Language Models: What Works and What Doesn't

Harikumar · 9 min read

Multi-hop reasoning -- answering questions that require chaining multiple facts -- remains one of the hardest open problems in NLP. We survey the benchmarks, the methods, the shortcuts, and the surprising failures that reveal how far we still have to go.

reasoningmulti hopcompositionality
RELATED
APR 10The Grokking Phenomenon: When Neural Networks Suddenly Generalize
6 min→
MAR 28Towards Verifiable AI: Formal Guarantees for Neural Network Outputs
7 min→
Reflqt.

Every claim, plotted. A research lab in public.

Research
All entriesPublicationsBlog
Explore
All demosAlgebraKnowledge Layer
Lab
About · Open letterContactGitHub
© 2026 REFLQT. LABSv2026.04 · plotted