{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:AAL3TGYGJ67GUGP3GBQCT37XHY","short_pith_number":"pith:AAL3TGYG","schema_version":"1.0","canonical_sha256":"0017b99b064fbe6a19fb306029eff73e1d470b3835f8e0befb836b1f3d200d42","source":{"kind":"arxiv","id":"2605.22005","version":1},"attestation_state":"computed","paper":{"title":"Check Your LLM's Secret Dictionary! Five Lines of Code Reveal What Your LLM Learned (Including What It Shouldn't Have)","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Hisashi Miyashita","submitted_at":"2026-05-21T05:02:51Z","abstract_excerpt":"We show that singular value decomposition of the lm_head} weight matrix of a transformer-based large language model -- requiring only five lines of PyTorch and no model inference -- reveals interpretable semantic subspaces directly from the model weights. Each left singular vector identifies the vocabulary tokens most readily selected when the hidden state aligns with the corresponding singular direction; inspecting these clusters exposes the model's training data composition and curation philosophy.\n  Analysing GPT-OSS-120B, Gemma-2-2B, and Qwen2.5-1.5B, we find that singular value spectra an"},"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":"2605.22005","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-05-21T05:02:51Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"7a8bc519d89a91e14e9b67d12b1c278469220ace1e586254392c9d3f628a8913","abstract_canon_sha256":"d2fd328238ba46a4663ec17642521a897116180df73e8322b13d7cfb5b14c708"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-22T01:04:20.037474Z","signature_b64":"HMHmA9iFfL2ay7Mj2EC61pNfR+m/HTc4rsvcRCub+VcoM5SV3YS6erHmPbyDtIMUmE0ToKbozcd1cYDDbHGaCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"0017b99b064fbe6a19fb306029eff73e1d470b3835f8e0befb836b1f3d200d42","last_reissued_at":"2026-05-22T01:04:20.036683Z","signature_status":"signed_v1","first_computed_at":"2026-05-22T01:04:20.036683Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Check Your LLM's Secret Dictionary! Five Lines of Code Reveal What Your LLM Learned (Including What It Shouldn't Have)","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Hisashi Miyashita","submitted_at":"2026-05-21T05:02:51Z","abstract_excerpt":"We show that singular value decomposition of the lm_head} weight matrix of a transformer-based large language model -- requiring only five lines of PyTorch and no model inference -- reveals interpretable semantic subspaces directly from the model weights. Each left singular vector identifies the vocabulary tokens most readily selected when the hidden state aligns with the corresponding singular direction; inspecting these clusters exposes the model's training data composition and curation philosophy.\n  Analysing GPT-OSS-120B, Gemma-2-2B, and Qwen2.5-1.5B, we find that singular value spectra an"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.22005","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/2605.22005/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":"2605.22005","created_at":"2026-05-22T01:04:20.036823+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.22005v1","created_at":"2026-05-22T01:04:20.036823+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.22005","created_at":"2026-05-22T01:04:20.036823+00:00"},{"alias_kind":"pith_short_12","alias_value":"AAL3TGYGJ67G","created_at":"2026-05-22T01:04:20.036823+00:00"},{"alias_kind":"pith_short_16","alias_value":"AAL3TGYGJ67GUGP3","created_at":"2026-05-22T01:04:20.036823+00:00"},{"alias_kind":"pith_short_8","alias_value":"AAL3TGYG","created_at":"2026-05-22T01:04:20.036823+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/AAL3TGYGJ67GUGP3GBQCT37XHY","json":"https://pith.science/pith/AAL3TGYGJ67GUGP3GBQCT37XHY.json","graph_json":"https://pith.science/api/pith-number/AAL3TGYGJ67GUGP3GBQCT37XHY/graph.json","events_json":"https://pith.science/api/pith-number/AAL3TGYGJ67GUGP3GBQCT37XHY/events.json","paper":"https://pith.science/paper/AAL3TGYG"},"agent_actions":{"view_html":"https://pith.science/pith/AAL3TGYGJ67GUGP3GBQCT37XHY","download_json":"https://pith.science/pith/AAL3TGYGJ67GUGP3GBQCT37XHY.json","view_paper":"https://pith.science/paper/AAL3TGYG","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.22005&json=true","fetch_graph":"https://pith.science/api/pith-number/AAL3TGYGJ67GUGP3GBQCT37XHY/graph.json","fetch_events":"https://pith.science/api/pith-number/AAL3TGYGJ67GUGP3GBQCT37XHY/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/AAL3TGYGJ67GUGP3GBQCT37XHY/action/timestamp_anchor","attest_storage":"https://pith.science/pith/AAL3TGYGJ67GUGP3GBQCT37XHY/action/storage_attestation","attest_author":"https://pith.science/pith/AAL3TGYGJ67GUGP3GBQCT37XHY/action/author_attestation","sign_citation":"https://pith.science/pith/AAL3TGYGJ67GUGP3GBQCT37XHY/action/citation_signature","submit_replication":"https://pith.science/pith/AAL3TGYGJ67GUGP3GBQCT37XHY/action/replication_record"}},"created_at":"2026-05-22T01:04:20.036823+00:00","updated_at":"2026-05-22T01:04:20.036823+00:00"}