{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:AZZECWFCZG6XEHZVSR2WCPBQX2","short_pith_number":"pith:AZZECWFC","schema_version":"1.0","canonical_sha256":"06724158a2c9bd721f359475613c30be9f7a800097c38b115614724c9a1c27d6","source":{"kind":"arxiv","id":"2307.09458","version":3},"attestation_state":"computed","paper":{"title":"Does Circuit Analysis Interpretability Scale? Evidence from Multiple Choice Capabilities in Chinchilla","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Geoffrey Irving, J\\'anos Kram\\'ar, Matthew Rahtz, Neel Nanda, Rohin Shah, Tom Lieberum, Vladimir Mikulik","submitted_at":"2023-07-18T17:39:04Z","abstract_excerpt":"\\emph{Circuit analysis} is a promising technique for understanding the internal mechanisms of language models. However, existing analyses are done in small models far from the state of the art. To address this, we present a case study of circuit analysis in the 70B Chinchilla model, aiming to test the scalability of circuit analysis. In particular, we study multiple-choice question answering, and investigate Chinchilla's capability to identify the correct answer \\emph{label} given knowledge of the correct answer \\emph{text}. We find that the existing techniques of logit attribution, attention "},"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":"2307.09458","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2023-07-18T17:39:04Z","cross_cats_sorted":[],"title_canon_sha256":"448d1d0eebf059e449da4335a41466adfe580d71cd859ea889489a1d979e2011","abstract_canon_sha256":"363fcc16b896b5bc7dad596569d2bd4acc7ce42d9a57a27d4cc81d34116bb275"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T06:33:47.596060Z","signature_b64":"485hYi3Imgfz4sicqwokE03Q61jM90moUTqZ4MNfcugmtu4UkXMevlYbamTOCrMCo6erBpF2I7dZxyQ6HWHlCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"06724158a2c9bd721f359475613c30be9f7a800097c38b115614724c9a1c27d6","last_reissued_at":"2026-07-05T06:33:47.595598Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T06:33:47.595598Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Does Circuit Analysis Interpretability Scale? Evidence from Multiple Choice Capabilities in Chinchilla","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Geoffrey Irving, J\\'anos Kram\\'ar, Matthew Rahtz, Neel Nanda, Rohin Shah, Tom Lieberum, Vladimir Mikulik","submitted_at":"2023-07-18T17:39:04Z","abstract_excerpt":"\\emph{Circuit analysis} is a promising technique for understanding the internal mechanisms of language models. However, existing analyses are done in small models far from the state of the art. To address this, we present a case study of circuit analysis in the 70B Chinchilla model, aiming to test the scalability of circuit analysis. In particular, we study multiple-choice question answering, and investigate Chinchilla's capability to identify the correct answer \\emph{label} given knowledge of the correct answer \\emph{text}. We find that the existing techniques of logit attribution, attention "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2307.09458","kind":"arxiv","version":3},"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/2307.09458/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":"2307.09458","created_at":"2026-07-05T06:33:47.595659+00:00"},{"alias_kind":"arxiv_version","alias_value":"2307.09458v3","created_at":"2026-07-05T06:33:47.595659+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2307.09458","created_at":"2026-07-05T06:33:47.595659+00:00"},{"alias_kind":"pith_short_12","alias_value":"AZZECWFCZG6X","created_at":"2026-07-05T06:33:47.595659+00:00"},{"alias_kind":"pith_short_16","alias_value":"AZZECWFCZG6XEHZV","created_at":"2026-07-05T06:33:47.595659+00:00"},{"alias_kind":"pith_short_8","alias_value":"AZZECWFC","created_at":"2026-07-05T06:33:47.595659+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":14,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.12818","citing_title":"Localizing Anchoring Pathways in Language Models","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2606.08644","citing_title":"A retrieval conditioned rebinding circuit for dynamic entity tracking in large language models","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2606.06267","citing_title":"Many Circuits, One Mechanism: Input Variation and Evaluation Granularity in Circuit Discovery","ref_index":50,"is_internal_anchor":false},{"citing_arxiv_id":"2606.05378","citing_title":"Pattern Selectivity is Not Task-Causal Structure: A Cross-Architecture Mechanistic Study of Composed-Task Circuits in 1B-Class Language Models","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2606.02378","citing_title":"When Do Attention Circuits Form? Developmental Trajectories of Capability and Attention-Sink Emergence Across Three 1B-ClassArchitectures","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2509.24164","citing_title":"Localizing Task Recognition and Task Learning in In-Context Learning via Attention Head Analysis","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2309.16042","citing_title":"Towards Best Practices of Activation Patching in Language Models: Metrics and Methods","ref_index":98,"is_internal_anchor":false},{"citing_arxiv_id":"2404.15255","citing_title":"How to use and interpret activation patching","ref_index":17,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12809","citing_title":"Correcting Influence: Unboxing LLM Outputs with Orthogonal Latent Spaces","ref_index":51,"is_internal_anchor":false},{"citing_arxiv_id":"2604.03764","citing_title":"Automated Attention Pattern Discovery at Scale in Large Language Models","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2605.09314","citing_title":"How LLMs Are Persuaded: A Few Attention Heads, Rerouted","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08853","citing_title":"Architecture, Not Scale: Circuit Localization in Large Language Models","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2605.05653","citing_title":"Negative Before Positive: Asymmetric Valence Processing in Large Language Models","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2604.13694","citing_title":"Weight Patching: Toward Source-Level Mechanistic Localization in LLMs","ref_index":46,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/AZZECWFCZG6XEHZVSR2WCPBQX2","json":"https://pith.science/pith/AZZECWFCZG6XEHZVSR2WCPBQX2.json","graph_json":"https://pith.science/api/pith-number/AZZECWFCZG6XEHZVSR2WCPBQX2/graph.json","events_json":"https://pith.science/api/pith-number/AZZECWFCZG6XEHZVSR2WCPBQX2/events.json","paper":"https://pith.science/paper/AZZECWFC"},"agent_actions":{"view_html":"https://pith.science/pith/AZZECWFCZG6XEHZVSR2WCPBQX2","download_json":"https://pith.science/pith/AZZECWFCZG6XEHZVSR2WCPBQX2.json","view_paper":"https://pith.science/paper/AZZECWFC","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2307.09458&json=true","fetch_graph":"https://pith.science/api/pith-number/AZZECWFCZG6XEHZVSR2WCPBQX2/graph.json","fetch_events":"https://pith.science/api/pith-number/AZZECWFCZG6XEHZVSR2WCPBQX2/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/AZZECWFCZG6XEHZVSR2WCPBQX2/action/timestamp_anchor","attest_storage":"https://pith.science/pith/AZZECWFCZG6XEHZVSR2WCPBQX2/action/storage_attestation","attest_author":"https://pith.science/pith/AZZECWFCZG6XEHZVSR2WCPBQX2/action/author_attestation","sign_citation":"https://pith.science/pith/AZZECWFCZG6XEHZVSR2WCPBQX2/action/citation_signature","submit_replication":"https://pith.science/pith/AZZECWFCZG6XEHZVSR2WCPBQX2/action/replication_record"}},"created_at":"2026-07-05T06:33:47.595659+00:00","updated_at":"2026-07-05T06:33:47.595659+00:00"}