{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:IPEZYXJUERVJPP5STAI3XFBHH7","short_pith_number":"pith:IPEZYXJU","schema_version":"1.0","canonical_sha256":"43c99c5d34246a97bfb29811bb94273fd9c70f61e7c0955cabbbaebfa08c89d4","source":{"kind":"arxiv","id":"2410.19750","version":2},"attestation_state":"computed","paper":{"title":"The Geometry of Concepts: Sparse Autoencoder Feature Structure","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"q-bio.NC","authors_text":"David D. Baek, Eric J. Michaud, Joshua Engels, Max Tegmark, Xiaoqing Sun, Yuxiao Li","submitted_at":"2024-10-10T17:58:47Z","abstract_excerpt":"Sparse autoencoders have recently produced dictionaries of high-dimensional vectors corresponding to the universe of concepts represented by large language models. We find that this concept universe has interesting structure at three levels: 1) The \"atomic\" small-scale structure contains \"crystals\" whose faces are parallelograms or trapezoids, generalizing well-known examples such as (man-woman-king-queen). We find that the quality of such parallelograms and associated function vectors improves greatly when projecting out global distractor directions such as word length, which is efficiently d"},"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":"2410.19750","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"q-bio.NC","submitted_at":"2024-10-10T17:58:47Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"b8643e889dcd185d6ab376e5fb94fe15a6ef4cc4143d8a0ab59f6bf59bb8d3cd","abstract_canon_sha256":"c8879b1400c2c57cc61a4af90031d45492fd41045e990a495ba086d5b069c5c4"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:41:49.815895Z","signature_b64":"06KqETLt+uYem2cGuTqJpQgdmQaBMFiYZ/4nEBomK8h9YcSiUuMK6EW0HgQgrVZbcrPULnsm6NNoUA9aO7OQDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"43c99c5d34246a97bfb29811bb94273fd9c70f61e7c0955cabbbaebfa08c89d4","last_reissued_at":"2026-07-05T10:41:49.815401Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:41:49.815401Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"The Geometry of Concepts: Sparse Autoencoder Feature Structure","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"q-bio.NC","authors_text":"David D. Baek, Eric J. Michaud, Joshua Engels, Max Tegmark, Xiaoqing Sun, Yuxiao Li","submitted_at":"2024-10-10T17:58:47Z","abstract_excerpt":"Sparse autoencoders have recently produced dictionaries of high-dimensional vectors corresponding to the universe of concepts represented by large language models. We find that this concept universe has interesting structure at three levels: 1) The \"atomic\" small-scale structure contains \"crystals\" whose faces are parallelograms or trapezoids, generalizing well-known examples such as (man-woman-king-queen). We find that the quality of such parallelograms and associated function vectors improves greatly when projecting out global distractor directions such as word length, which is efficiently d"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2410.19750","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/2410.19750/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":"2410.19750","created_at":"2026-07-05T10:41:49.815461+00:00"},{"alias_kind":"arxiv_version","alias_value":"2410.19750v2","created_at":"2026-07-05T10:41:49.815461+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2410.19750","created_at":"2026-07-05T10:41:49.815461+00:00"},{"alias_kind":"pith_short_12","alias_value":"IPEZYXJUERVJ","created_at":"2026-07-05T10:41:49.815461+00:00"},{"alias_kind":"pith_short_16","alias_value":"IPEZYXJUERVJPP5S","created_at":"2026-07-05T10:41:49.815461+00:00"},{"alias_kind":"pith_short_8","alias_value":"IPEZYXJU","created_at":"2026-07-05T10:41:49.815461+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.26987","citing_title":"Where Do Models Find Happiness? Emotion Vectors in Open-Source LLMs","ref_index":1,"is_internal_anchor":false},{"citing_arxiv_id":"2605.09887","citing_title":"The Geometric Wall: Manifold Structure Predicts Layerwise Sparse Autoencoder Scaling Laws","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06494","citing_title":"From Token Lists to Graph Motifs: Weisfeiler-Lehman Analysis of Sparse Autoencoder Features","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2604.18756","citing_title":"Towards Understanding the Robustness of Sparse Autoencoders","ref_index":6,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/IPEZYXJUERVJPP5STAI3XFBHH7","json":"https://pith.science/pith/IPEZYXJUERVJPP5STAI3XFBHH7.json","graph_json":"https://pith.science/api/pith-number/IPEZYXJUERVJPP5STAI3XFBHH7/graph.json","events_json":"https://pith.science/api/pith-number/IPEZYXJUERVJPP5STAI3XFBHH7/events.json","paper":"https://pith.science/paper/IPEZYXJU"},"agent_actions":{"view_html":"https://pith.science/pith/IPEZYXJUERVJPP5STAI3XFBHH7","download_json":"https://pith.science/pith/IPEZYXJUERVJPP5STAI3XFBHH7.json","view_paper":"https://pith.science/paper/IPEZYXJU","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2410.19750&json=true","fetch_graph":"https://pith.science/api/pith-number/IPEZYXJUERVJPP5STAI3XFBHH7/graph.json","fetch_events":"https://pith.science/api/pith-number/IPEZYXJUERVJPP5STAI3XFBHH7/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/IPEZYXJUERVJPP5STAI3XFBHH7/action/timestamp_anchor","attest_storage":"https://pith.science/pith/IPEZYXJUERVJPP5STAI3XFBHH7/action/storage_attestation","attest_author":"https://pith.science/pith/IPEZYXJUERVJPP5STAI3XFBHH7/action/author_attestation","sign_citation":"https://pith.science/pith/IPEZYXJUERVJPP5STAI3XFBHH7/action/citation_signature","submit_replication":"https://pith.science/pith/IPEZYXJUERVJPP5STAI3XFBHH7/action/replication_record"}},"created_at":"2026-07-05T10:41:49.815461+00:00","updated_at":"2026-07-05T10:41:49.815461+00:00"}