{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:HFM7HJHD5BLJOHHUVMJM7MB2UI","short_pith_number":"pith:HFM7HJHD","schema_version":"1.0","canonical_sha256":"3959f3a4e3e856971cf4ab12cfb03aa21f6912b67e87dcee163814b0f74631bd","source":{"kind":"arxiv","id":"2307.03637","version":1},"attestation_state":"computed","paper":{"title":"Discovering Variable Binding Circuitry with Desiderata","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"David Bau, Max Nadeau, Nikhil Prakash, Tamar Rott Shaham, Xander Davies","submitted_at":"2023-07-07T14:51:30Z","abstract_excerpt":"Recent work has shown that computation in language models may be human-understandable, with successful efforts to localize and intervene on both single-unit features and input-output circuits. Here, we introduce an approach which extends causal mediation experiments to automatically identify model components responsible for performing a specific subtask by solely specifying a set of \\textit{desiderata}, or causal attributes of the model components executing that subtask. As a proof of concept, we apply our method to automatically discover shared \\textit{variable binding circuitry} in LLaMA-13B"},"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":"2307.03637","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2023-07-07T14:51:30Z","cross_cats_sorted":[],"title_canon_sha256":"dd1777e9ad0188e09afbdc1467fddc18a6691af8bfea96571e0ceed6e8d76c93","abstract_canon_sha256":"1db8115365e7c41afad8b02559b8e600d094e8d22e0216bfe81477c490813dab"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T06:28:45.524728Z","signature_b64":"uh6rxy8POGnj/DyqhGe99Aqc3nr0lrLbdDMIn0Q1W5u9DZk+vJdieJbkaPj0bT+8oCDFlLGxHok2onN+uVR4Cw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"3959f3a4e3e856971cf4ab12cfb03aa21f6912b67e87dcee163814b0f74631bd","last_reissued_at":"2026-07-05T06:28:45.524220Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T06:28:45.524220Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Discovering Variable Binding Circuitry with Desiderata","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"David Bau, Max Nadeau, Nikhil Prakash, Tamar Rott Shaham, Xander Davies","submitted_at":"2023-07-07T14:51:30Z","abstract_excerpt":"Recent work has shown that computation in language models may be human-understandable, with successful efforts to localize and intervene on both single-unit features and input-output circuits. Here, we introduce an approach which extends causal mediation experiments to automatically identify model components responsible for performing a specific subtask by solely specifying a set of \\textit{desiderata}, or causal attributes of the model components executing that subtask. As a proof of concept, we apply our method to automatically discover shared \\textit{variable binding circuitry} in LLaMA-13B"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2307.03637","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/2307.03637/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":"2307.03637","created_at":"2026-07-05T06:28:45.524279+00:00"},{"alias_kind":"arxiv_version","alias_value":"2307.03637v1","created_at":"2026-07-05T06:28:45.524279+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2307.03637","created_at":"2026-07-05T06:28:45.524279+00:00"},{"alias_kind":"pith_short_12","alias_value":"HFM7HJHD5BLJ","created_at":"2026-07-05T06:28:45.524279+00:00"},{"alias_kind":"pith_short_16","alias_value":"HFM7HJHD5BLJOHHU","created_at":"2026-07-05T06:28:45.524279+00:00"},{"alias_kind":"pith_short_8","alias_value":"HFM7HJHD","created_at":"2026-07-05T06:28:45.524279+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2309.16042","citing_title":"Towards Best Practices of Activation Patching in Language Models: Metrics and Methods","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12809","citing_title":"Correcting Influence: Unboxing LLM Outputs with Orthogonal Latent Spaces","ref_index":74,"is_internal_anchor":false},{"citing_arxiv_id":"2605.01148","citing_title":"Arithmetic in the Wild: Llama uses Base-10 Addition to Reason About Cyclic Concepts","ref_index":130,"is_internal_anchor":false},{"citing_arxiv_id":"2605.05115","citing_title":"Manifold Steering Reveals the Shared Geometry of Neural Network Representation and Behavior","ref_index":125,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/HFM7HJHD5BLJOHHUVMJM7MB2UI","json":"https://pith.science/pith/HFM7HJHD5BLJOHHUVMJM7MB2UI.json","graph_json":"https://pith.science/api/pith-number/HFM7HJHD5BLJOHHUVMJM7MB2UI/graph.json","events_json":"https://pith.science/api/pith-number/HFM7HJHD5BLJOHHUVMJM7MB2UI/events.json","paper":"https://pith.science/paper/HFM7HJHD"},"agent_actions":{"view_html":"https://pith.science/pith/HFM7HJHD5BLJOHHUVMJM7MB2UI","download_json":"https://pith.science/pith/HFM7HJHD5BLJOHHUVMJM7MB2UI.json","view_paper":"https://pith.science/paper/HFM7HJHD","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2307.03637&json=true","fetch_graph":"https://pith.science/api/pith-number/HFM7HJHD5BLJOHHUVMJM7MB2UI/graph.json","fetch_events":"https://pith.science/api/pith-number/HFM7HJHD5BLJOHHUVMJM7MB2UI/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/HFM7HJHD5BLJOHHUVMJM7MB2UI/action/timestamp_anchor","attest_storage":"https://pith.science/pith/HFM7HJHD5BLJOHHUVMJM7MB2UI/action/storage_attestation","attest_author":"https://pith.science/pith/HFM7HJHD5BLJOHHUVMJM7MB2UI/action/author_attestation","sign_citation":"https://pith.science/pith/HFM7HJHD5BLJOHHUVMJM7MB2UI/action/citation_signature","submit_replication":"https://pith.science/pith/HFM7HJHD5BLJOHHUVMJM7MB2UI/action/replication_record"}},"created_at":"2026-07-05T06:28:45.524279+00:00","updated_at":"2026-07-05T06:28:45.524279+00:00"}