{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:7TIXJEQRBZ6JMDREN46ZCFDTIK","short_pith_number":"pith:7TIXJEQR","schema_version":"1.0","canonical_sha256":"fcd17492110e7c960e246f3d9114734299cd0f84d7e24992af1aee370caf9f41","source":{"kind":"arxiv","id":"2410.13032","version":1},"attestation_state":"computed","paper":{"title":"Hypothesis Testing the Circuit Hypothesis in LLMs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG","stat.ML"],"primary_cat":"cs.AI","authors_text":"Achille Nazaret, Adri\\`a Garriga-Alonso, Andrew Jesson, Carolina Zheng, Claudia Shi, David M. Blei, Maggie Makar, Nicolas Beltran-Velez","submitted_at":"2024-10-16T20:45:29Z","abstract_excerpt":"Large language models (LLMs) demonstrate surprising capabilities, but we do not understand how they are implemented. One hypothesis suggests that these capabilities are primarily executed by small subnetworks within the LLM, known as circuits. But how can we evaluate this hypothesis? In this paper, we formalize a set of criteria that a circuit is hypothesized to meet and develop a suite of hypothesis tests to evaluate how well circuits satisfy them. The criteria focus on the extent to which the LLM's behavior is preserved, the degree of localization of this behavior, and whether the circuit is"},"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":"2410.13032","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2024-10-16T20:45:29Z","cross_cats_sorted":["cs.LG","stat.ML"],"title_canon_sha256":"cedc25652923fce07e2f1582e2b45831c56a4dfdfee811fb0dc45a50ff6d79a1","abstract_canon_sha256":"b248ce700fe7df36006ee0bc845c4a8a4f2e0a6e986f7050b05337530db46b35"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:21:54.386602Z","signature_b64":"y19KkUFIX6J+Lfw+9jjdydD6ASFHCI9kqOu0jEPH+qydvXtXP03kE3i24YtrZMFVD8wARd7fu2jJI7ae4vgnCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"fcd17492110e7c960e246f3d9114734299cd0f84d7e24992af1aee370caf9f41","last_reissued_at":"2026-07-05T09:21:54.386091Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:21:54.386091Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Hypothesis Testing the Circuit Hypothesis in LLMs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG","stat.ML"],"primary_cat":"cs.AI","authors_text":"Achille Nazaret, Adri\\`a Garriga-Alonso, Andrew Jesson, Carolina Zheng, Claudia Shi, David M. Blei, Maggie Makar, Nicolas Beltran-Velez","submitted_at":"2024-10-16T20:45:29Z","abstract_excerpt":"Large language models (LLMs) demonstrate surprising capabilities, but we do not understand how they are implemented. One hypothesis suggests that these capabilities are primarily executed by small subnetworks within the LLM, known as circuits. But how can we evaluate this hypothesis? In this paper, we formalize a set of criteria that a circuit is hypothesized to meet and develop a suite of hypothesis tests to evaluate how well circuits satisfy them. The criteria focus on the extent to which the LLM's behavior is preserved, the degree of localization of this behavior, and whether the circuit is"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2410.13032","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/2410.13032/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":"2410.13032","created_at":"2026-07-05T09:21:54.386154+00:00"},{"alias_kind":"arxiv_version","alias_value":"2410.13032v1","created_at":"2026-07-05T09:21:54.386154+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2410.13032","created_at":"2026-07-05T09:21:54.386154+00:00"},{"alias_kind":"pith_short_12","alias_value":"7TIXJEQRBZ6J","created_at":"2026-07-05T09:21:54.386154+00:00"},{"alias_kind":"pith_short_16","alias_value":"7TIXJEQRBZ6JMDRE","created_at":"2026-07-05T09:21:54.386154+00:00"},{"alias_kind":"pith_short_8","alias_value":"7TIXJEQR","created_at":"2026-07-05T09:21:54.386154+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.25092","citing_title":"How Modular Is a Frontier Mixture-of-Experts? A Pre-registered Causal Test in Which Apparent Expert Modularity Mostly Dissolves","ref_index":9,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/7TIXJEQRBZ6JMDREN46ZCFDTIK","json":"https://pith.science/pith/7TIXJEQRBZ6JMDREN46ZCFDTIK.json","graph_json":"https://pith.science/api/pith-number/7TIXJEQRBZ6JMDREN46ZCFDTIK/graph.json","events_json":"https://pith.science/api/pith-number/7TIXJEQRBZ6JMDREN46ZCFDTIK/events.json","paper":"https://pith.science/paper/7TIXJEQR"},"agent_actions":{"view_html":"https://pith.science/pith/7TIXJEQRBZ6JMDREN46ZCFDTIK","download_json":"https://pith.science/pith/7TIXJEQRBZ6JMDREN46ZCFDTIK.json","view_paper":"https://pith.science/paper/7TIXJEQR","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2410.13032&json=true","fetch_graph":"https://pith.science/api/pith-number/7TIXJEQRBZ6JMDREN46ZCFDTIK/graph.json","fetch_events":"https://pith.science/api/pith-number/7TIXJEQRBZ6JMDREN46ZCFDTIK/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/7TIXJEQRBZ6JMDREN46ZCFDTIK/action/timestamp_anchor","attest_storage":"https://pith.science/pith/7TIXJEQRBZ6JMDREN46ZCFDTIK/action/storage_attestation","attest_author":"https://pith.science/pith/7TIXJEQRBZ6JMDREN46ZCFDTIK/action/author_attestation","sign_citation":"https://pith.science/pith/7TIXJEQRBZ6JMDREN46ZCFDTIK/action/citation_signature","submit_replication":"https://pith.science/pith/7TIXJEQRBZ6JMDREN46ZCFDTIK/action/replication_record"}},"created_at":"2026-07-05T09:21:54.386154+00:00","updated_at":"2026-07-05T09:21:54.386154+00:00"}