{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:U3LJPS4YTCECQ6IKF5SZ4MGLQZ","short_pith_number":"pith:U3LJPS4Y","schema_version":"1.0","canonical_sha256":"a6d697cb98988828790a2f659e30cb8658f01ac42b9d5e41c3708e09c72010a6","source":{"kind":"arxiv","id":"2310.14491","version":1},"attestation_state":"computed","paper":{"title":"Towards a Mechanistic Interpretation of Multi-Step Reasoning Capabilities of Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Alessandro Stolfo, Antoine Bosselut, Guangtao Zeng, Jiaoda Li, Mrinmaya Sachan, Wangchunshu Zhou, Yifan Hou, Yu Fei","submitted_at":"2023-10-23T01:47:29Z","abstract_excerpt":"Recent work has shown that language models (LMs) have strong multi-step (i.e., procedural) reasoning capabilities. However, it is unclear whether LMs perform these tasks by cheating with answers memorized from pretraining corpus, or, via a multi-step reasoning mechanism. In this paper, we try to answer this question by exploring a mechanistic interpretation of LMs for multi-step reasoning tasks. Concretely, we hypothesize that the LM implicitly embeds a reasoning tree resembling the correct reasoning process within it. We test this hypothesis by introducing a new probing approach (called Mecha"},"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":"2310.14491","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2023-10-23T01:47:29Z","cross_cats_sorted":[],"title_canon_sha256":"454ba7173c52ed4f4d0013a3ed3d1338ec7f0951d24519bb8753047262d30fac","abstract_canon_sha256":"14464120303338eb6ef7dd2e604fadec63ef7112b3d224e880d79d127d337a86"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:03:45.099582Z","signature_b64":"ufuwMjYdYXAGcjueg9Zh68u4eWusP3E/Rh1x49VXF7sTMa3cv3bRcO4dyVY10sGIKD38AQi5N0ryMFfLoSFJDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a6d697cb98988828790a2f659e30cb8658f01ac42b9d5e41c3708e09c72010a6","last_reissued_at":"2026-07-05T07:03:45.099034Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:03:45.099034Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Towards a Mechanistic Interpretation of Multi-Step Reasoning Capabilities of Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Alessandro Stolfo, Antoine Bosselut, Guangtao Zeng, Jiaoda Li, Mrinmaya Sachan, Wangchunshu Zhou, Yifan Hou, Yu Fei","submitted_at":"2023-10-23T01:47:29Z","abstract_excerpt":"Recent work has shown that language models (LMs) have strong multi-step (i.e., procedural) reasoning capabilities. However, it is unclear whether LMs perform these tasks by cheating with answers memorized from pretraining corpus, or, via a multi-step reasoning mechanism. In this paper, we try to answer this question by exploring a mechanistic interpretation of LMs for multi-step reasoning tasks. Concretely, we hypothesize that the LM implicitly embeds a reasoning tree resembling the correct reasoning process within it. We test this hypothesis by introducing a new probing approach (called Mecha"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2310.14491","kind":"arxiv","version":1},"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/2310.14491/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":"2310.14491","created_at":"2026-07-05T07:03:45.099091+00:00"},{"alias_kind":"arxiv_version","alias_value":"2310.14491v1","created_at":"2026-07-05T07:03:45.099091+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2310.14491","created_at":"2026-07-05T07:03:45.099091+00:00"},{"alias_kind":"pith_short_12","alias_value":"U3LJPS4YTCEC","created_at":"2026-07-05T07:03:45.099091+00:00"},{"alias_kind":"pith_short_16","alias_value":"U3LJPS4YTCECQ6IK","created_at":"2026-07-05T07:03:45.099091+00:00"},{"alias_kind":"pith_short_8","alias_value":"U3LJPS4Y","created_at":"2026-07-05T07:03:45.099091+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/U3LJPS4YTCECQ6IKF5SZ4MGLQZ","json":"https://pith.science/pith/U3LJPS4YTCECQ6IKF5SZ4MGLQZ.json","graph_json":"https://pith.science/api/pith-number/U3LJPS4YTCECQ6IKF5SZ4MGLQZ/graph.json","events_json":"https://pith.science/api/pith-number/U3LJPS4YTCECQ6IKF5SZ4MGLQZ/events.json","paper":"https://pith.science/paper/U3LJPS4Y"},"agent_actions":{"view_html":"https://pith.science/pith/U3LJPS4YTCECQ6IKF5SZ4MGLQZ","download_json":"https://pith.science/pith/U3LJPS4YTCECQ6IKF5SZ4MGLQZ.json","view_paper":"https://pith.science/paper/U3LJPS4Y","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2310.14491&json=true","fetch_graph":"https://pith.science/api/pith-number/U3LJPS4YTCECQ6IKF5SZ4MGLQZ/graph.json","fetch_events":"https://pith.science/api/pith-number/U3LJPS4YTCECQ6IKF5SZ4MGLQZ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/U3LJPS4YTCECQ6IKF5SZ4MGLQZ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/U3LJPS4YTCECQ6IKF5SZ4MGLQZ/action/storage_attestation","attest_author":"https://pith.science/pith/U3LJPS4YTCECQ6IKF5SZ4MGLQZ/action/author_attestation","sign_citation":"https://pith.science/pith/U3LJPS4YTCECQ6IKF5SZ4MGLQZ/action/citation_signature","submit_replication":"https://pith.science/pith/U3LJPS4YTCECQ6IKF5SZ4MGLQZ/action/replication_record"}},"created_at":"2026-07-05T07:03:45.099091+00:00","updated_at":"2026-07-05T07:03:45.099091+00:00"}