{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:CRSK6HV6IWU7JTCASXEKU2OZV5","short_pith_number":"pith:CRSK6HV6","schema_version":"1.0","canonical_sha256":"1464af1ebe45a9f4cc4095c8aa69d9af7032b0e5006773df402c69194e923341","source":{"kind":"arxiv","id":"2402.16837","version":2},"attestation_state":"computed","paper":{"title":"Do Large Language Models Latently Perform Multi-Hop Reasoning?","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Elena Gribovskaya, Mor Geva, Nora Kassner, Sebastian Riedel, Sohee Yang","submitted_at":"2024-02-26T18:57:54Z","abstract_excerpt":"We study whether Large Language Models (LLMs) latently perform multi-hop reasoning with complex prompts such as \"The mother of the singer of 'Superstition' is\". We look for evidence of a latent reasoning pathway where an LLM (1) latently identifies \"the singer of 'Superstition'\" as Stevie Wonder, the bridge entity, and (2) uses its knowledge of Stevie Wonder's mother to complete the prompt. We analyze these two hops individually and consider their co-occurrence as indicative of latent multi-hop reasoning. For the first hop, we test if changing the prompt to indirectly mention the bridge entity"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2402.16837","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-02-26T18:57:54Z","cross_cats_sorted":[],"title_canon_sha256":"520fe99406a70e93c5ed6119c96276c0ccf2134d780543044b1fb7425d992ea5","abstract_canon_sha256":"81c26de4af3768efcdf38e002575f3c956a01844e5edba45c43a8f915f45157b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:13:07.737801Z","signature_b64":"6ZFgHfMCvA1vAewWjVr7pBmJI1yIZJyFtOo92zLQ6z3oSvCU+OWpVjCNQVA0HALZaVWoXBhHMQqIkBITbsJcAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"1464af1ebe45a9f4cc4095c8aa69d9af7032b0e5006773df402c69194e923341","last_reissued_at":"2026-07-05T11:13:07.737256Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:13:07.737256Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Do Large Language Models Latently Perform Multi-Hop Reasoning?","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Elena Gribovskaya, Mor Geva, Nora Kassner, Sebastian Riedel, Sohee Yang","submitted_at":"2024-02-26T18:57:54Z","abstract_excerpt":"We study whether Large Language Models (LLMs) latently perform multi-hop reasoning with complex prompts such as \"The mother of the singer of 'Superstition' is\". We look for evidence of a latent reasoning pathway where an LLM (1) latently identifies \"the singer of 'Superstition'\" as Stevie Wonder, the bridge entity, and (2) uses its knowledge of Stevie Wonder's mother to complete the prompt. We analyze these two hops individually and consider their co-occurrence as indicative of latent multi-hop reasoning. For the first hop, we test if changing the prompt to indirectly mention the bridge entity"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2402.16837","kind":"arxiv","version":2},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2402.16837/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2402.16837","created_at":"2026-07-05T11:13:07.737331+00:00"},{"alias_kind":"arxiv_version","alias_value":"2402.16837v2","created_at":"2026-07-05T11:13:07.737331+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2402.16837","created_at":"2026-07-05T11:13:07.737331+00:00"},{"alias_kind":"pith_short_12","alias_value":"CRSK6HV6IWU7","created_at":"2026-07-05T11:13:07.737331+00:00"},{"alias_kind":"pith_short_16","alias_value":"CRSK6HV6IWU7JTCA","created_at":"2026-07-05T11:13:07.737331+00:00"},{"alias_kind":"pith_short_8","alias_value":"CRSK6HV6","created_at":"2026-07-05T11:13:07.737331+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":16,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.07157","citing_title":"Think Fast: Estimating No-CoT Task-Completion Time Horizons of Frontier AI Models","ref_index":31,"is_internal_anchor":false},{"citing_arxiv_id":"2607.00341","citing_title":"DiscoLoop: Looping Discrete Embeddings and Continuous Hidden States for Multi-hop Reasoning","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2606.07157","citing_title":"Think Fast: Estimating No-CoT Task-Completion Time Horizons of Frontier AI Models","ref_index":31,"is_internal_anchor":false},{"citing_arxiv_id":"2501.19201","citing_title":"Efficient Reasoning with Hidden Thinking","ref_index":25,"is_internal_anchor":false},{"citing_arxiv_id":"2506.04289","citing_title":"Relational reasoning and inductive bias in transformers and large language models","ref_index":24,"is_internal_anchor":false},{"citing_arxiv_id":"2510.24941","citing_title":"Can Aha Moments Be Fake? Towards Quantifying Decorative and True Thinking in Chain-of-Thought","ref_index":33,"is_internal_anchor":false},{"citing_arxiv_id":"2409.12917","citing_title":"Training Language Models to Self-Correct via Reinforcement Learning","ref_index":48,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08221","citing_title":"NoisyCoconut: Counterfactual Consensus via Latent Space Reasoning","ref_index":43,"is_internal_anchor":false},{"citing_arxiv_id":"2604.15529","citing_title":"LACE: Lattice Attention for Cross-thread Exploration","ref_index":46,"is_internal_anchor":false},{"citing_arxiv_id":"2604.22951","citing_title":"The Power of Power Law: Asymmetry Enables Compositional Reasoning","ref_index":56,"is_internal_anchor":false},{"citing_arxiv_id":"2412.06769","citing_title":"Training Large Language Models to Reason in a Continuous Latent Space","ref_index":32,"is_internal_anchor":false},{"citing_arxiv_id":"2604.15529","citing_title":"LACE: Lattice Attention for Cross-thread Exploration","ref_index":46,"is_internal_anchor":false},{"citing_arxiv_id":"2604.08299","citing_title":"SeLaR: Selective Latent Reasoning in Large Language Models","ref_index":52,"is_internal_anchor":false},{"citing_arxiv_id":"2604.15529","citing_title":"LACE: Lattice Attention for Cross-thread Exploration","ref_index":46,"is_internal_anchor":false},{"citing_arxiv_id":"2604.17458","citing_title":"EHRAG: Bridging Semantic Gaps in Lightweight GraphRAG via Hybrid Hypergraph Construction and Retrieval","ref_index":172,"is_internal_anchor":false},{"citing_arxiv_id":"2604.21027","citing_title":"HypEHR: Hyperbolic Modeling of Electronic Health Records for Efficient Question Answering","ref_index":102,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/CRSK6HV6IWU7JTCASXEKU2OZV5","json":"https://pith.science/pith/CRSK6HV6IWU7JTCASXEKU2OZV5.json","graph_json":"https://pith.science/api/pith-number/CRSK6HV6IWU7JTCASXEKU2OZV5/graph.json","events_json":"https://pith.science/api/pith-number/CRSK6HV6IWU7JTCASXEKU2OZV5/events.json","paper":"https://pith.science/paper/CRSK6HV6"},"agent_actions":{"view_html":"https://pith.science/pith/CRSK6HV6IWU7JTCASXEKU2OZV5","download_json":"https://pith.science/pith/CRSK6HV6IWU7JTCASXEKU2OZV5.json","view_paper":"https://pith.science/paper/CRSK6HV6","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2402.16837&json=true","fetch_graph":"https://pith.science/api/pith-number/CRSK6HV6IWU7JTCASXEKU2OZV5/graph.json","fetch_events":"https://pith.science/api/pith-number/CRSK6HV6IWU7JTCASXEKU2OZV5/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/CRSK6HV6IWU7JTCASXEKU2OZV5/action/timestamp_anchor","attest_storage":"https://pith.science/pith/CRSK6HV6IWU7JTCASXEKU2OZV5/action/storage_attestation","attest_author":"https://pith.science/pith/CRSK6HV6IWU7JTCASXEKU2OZV5/action/author_attestation","sign_citation":"https://pith.science/pith/CRSK6HV6IWU7JTCASXEKU2OZV5/action/citation_signature","submit_replication":"https://pith.science/pith/CRSK6HV6IWU7JTCASXEKU2OZV5/action/replication_record"}},"created_at":"2026-07-05T11:13:07.737331+00:00","updated_at":"2026-07-05T11:13:07.737331+00:00"}