{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:QBCNCIADI3DMLHXXQR72OWKMUP","short_pith_number":"pith:QBCNCIAD","schema_version":"1.0","canonical_sha256":"8044d1200346c6c59ef7847fa7594ca3dbc7850eedb0b2990238c8156ce28479","source":{"kind":"arxiv","id":"2209.02535","version":3},"attestation_state":"computed","paper":{"title":"Analyzing Transformers in Embedding Space","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Ankit Gupta, Guy Dar, Jonathan Berant, Mor Geva","submitted_at":"2022-09-06T14:36:57Z","abstract_excerpt":"Understanding Transformer-based models has attracted significant attention, as they lie at the heart of recent technological advances across machine learning. While most interpretability methods rely on running models over inputs, recent work has shown that a zero-pass approach, where parameters are interpreted directly without a forward/backward pass is feasible for some Transformer parameters, and for two-layer attention networks. In this work, we present a theoretical analysis where all parameters of a trained Transformer are interpreted by projecting them into the embedding space, that 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":"2209.02535","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2022-09-06T14:36:57Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"7681ef1446a6dfc8ce849a756f0d553fffd1083f2fb31f9eb93b87dc74476c34","abstract_canon_sha256":"5846a6e2107810a5366b7c6284adabf3f83add9d9d7834b197cdb2789530d00f"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:27:41.323260Z","signature_b64":"EaUWZO5WkQ54/AdwQREDn6V5vMhMXFpxg7mC7397UwfGJ9sS1o/OWilqpGRnpVusePrj8eXv23XssPB9ijHxAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"8044d1200346c6c59ef7847fa7594ca3dbc7850eedb0b2990238c8156ce28479","last_reissued_at":"2026-07-05T07:27:41.322680Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:27:41.322680Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Analyzing Transformers in Embedding Space","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Ankit Gupta, Guy Dar, Jonathan Berant, Mor Geva","submitted_at":"2022-09-06T14:36:57Z","abstract_excerpt":"Understanding Transformer-based models has attracted significant attention, as they lie at the heart of recent technological advances across machine learning. While most interpretability methods rely on running models over inputs, recent work has shown that a zero-pass approach, where parameters are interpreted directly without a forward/backward pass is feasible for some Transformer parameters, and for two-layer attention networks. In this work, we present a theoretical analysis where all parameters of a trained Transformer are interpreted by projecting them into the embedding space, that is,"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2209.02535","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/2209.02535/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":"2209.02535","created_at":"2026-07-05T07:27:41.322752+00:00"},{"alias_kind":"arxiv_version","alias_value":"2209.02535v3","created_at":"2026-07-05T07:27:41.322752+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2209.02535","created_at":"2026-07-05T07:27:41.322752+00:00"},{"alias_kind":"pith_short_12","alias_value":"QBCNCIADI3DM","created_at":"2026-07-05T07:27:41.322752+00:00"},{"alias_kind":"pith_short_16","alias_value":"QBCNCIADI3DMLHXX","created_at":"2026-07-05T07:27:41.322752+00:00"},{"alias_kind":"pith_short_8","alias_value":"QBCNCIAD","created_at":"2026-07-05T07:27:41.322752+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":10,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.06445","citing_title":"Analysis-by-Proxy: Localization Signals in VLMs Operating as Condition Encoders","ref_index":3,"is_internal_anchor":true},{"citing_arxiv_id":"2607.01571","citing_title":"Geometric Signatures of Reasoning: A Spectral Perspective on Task Hardness","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2605.27970","citing_title":"Geometry of Human Perceptual Domains Emerges Transiently in LLM Representations","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2604.01404","citing_title":"Friends and Grandmothers in Silico: Localizing Entity Cells in Language Models","ref_index":1,"is_internal_anchor":false},{"citing_arxiv_id":"2404.16014","citing_title":"Improving Dictionary Learning with Gated Sparse Autoencoders","ref_index":158,"is_internal_anchor":false},{"citing_arxiv_id":"2210.07229","citing_title":"Mass-Editing Memory in a Transformer","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12809","citing_title":"Correcting Influence: Unboxing LLM Outputs with Orthogonal Latent Spaces","ref_index":233,"is_internal_anchor":false},{"citing_arxiv_id":"2211.00593","citing_title":"Interpretability in the Wild: a Circuit for Indirect Object Identification in GPT-2 small","ref_index":50,"is_internal_anchor":false},{"citing_arxiv_id":"2303.08112","citing_title":"Eliciting Latent Predictions from Transformers with the Tuned Lens","ref_index":26,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08934","citing_title":"From Mechanistic to Compositional Interpretability","ref_index":21,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/QBCNCIADI3DMLHXXQR72OWKMUP","json":"https://pith.science/pith/QBCNCIADI3DMLHXXQR72OWKMUP.json","graph_json":"https://pith.science/api/pith-number/QBCNCIADI3DMLHXXQR72OWKMUP/graph.json","events_json":"https://pith.science/api/pith-number/QBCNCIADI3DMLHXXQR72OWKMUP/events.json","paper":"https://pith.science/paper/QBCNCIAD"},"agent_actions":{"view_html":"https://pith.science/pith/QBCNCIADI3DMLHXXQR72OWKMUP","download_json":"https://pith.science/pith/QBCNCIADI3DMLHXXQR72OWKMUP.json","view_paper":"https://pith.science/paper/QBCNCIAD","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2209.02535&json=true","fetch_graph":"https://pith.science/api/pith-number/QBCNCIADI3DMLHXXQR72OWKMUP/graph.json","fetch_events":"https://pith.science/api/pith-number/QBCNCIADI3DMLHXXQR72OWKMUP/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/QBCNCIADI3DMLHXXQR72OWKMUP/action/timestamp_anchor","attest_storage":"https://pith.science/pith/QBCNCIADI3DMLHXXQR72OWKMUP/action/storage_attestation","attest_author":"https://pith.science/pith/QBCNCIADI3DMLHXXQR72OWKMUP/action/author_attestation","sign_citation":"https://pith.science/pith/QBCNCIADI3DMLHXXQR72OWKMUP/action/citation_signature","submit_replication":"https://pith.science/pith/QBCNCIADI3DMLHXXQR72OWKMUP/action/replication_record"}},"created_at":"2026-07-05T07:27:41.322752+00:00","updated_at":"2026-07-05T07:27:41.322752+00:00"}