{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:UPRZD74KTMQLKHXE7SFDIYXDAN","short_pith_number":"pith:UPRZD74K","schema_version":"1.0","canonical_sha256":"a3e391ff8a9b20b51ee4fc8a3462e30347c67972f51c25717df3c037241949c5","source":{"kind":"arxiv","id":"2506.21468","version":1},"attestation_state":"computed","paper":{"title":"TopK Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Benjamin Heinzerling, Kentaro Inui, Ryosuke Takahashi, Tatsuro Inaba","submitted_at":"2025-06-26T16:56:43Z","abstract_excerpt":"Sparse autoencoders (SAEs) have become an important tool for analyzing and interpreting the activation space of transformer-based language models (LMs). However, SAEs suffer several shortcomings that diminish their utility and internal validity. Since SAEs are trained post-hoc, it is unclear if the failure to discover a particular concept is a failure on the SAE's side or due to the underlying LM not representing this concept. This problem is exacerbated by training conditions and architecture choices affecting which features an SAE learns. When tracing how LMs learn concepts during training, "},"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":"2506.21468","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-06-26T16:56:43Z","cross_cats_sorted":[],"title_canon_sha256":"dc38e9590648123a7b589dcf3a774e5c53447b3c58c38a583a475a7118ef76ec","abstract_canon_sha256":"edbdf4e5c4d339d2fc58240572683c373cdd6eea355c4d6dc0d4c0dc8bbc544b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:27:46.470716Z","signature_b64":"p3Txe3Cyd5Jdaj66EjlNuhByuGTxvLVzJFsvVu8QeJj8ZwBFkUEtiZu/wDj8taYXQYzX2/4CG0dPLTnI7GrsBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a3e391ff8a9b20b51ee4fc8a3462e30347c67972f51c25717df3c037241949c5","last_reissued_at":"2026-07-05T11:27:46.470242Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:27:46.470242Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"TopK Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Benjamin Heinzerling, Kentaro Inui, Ryosuke Takahashi, Tatsuro Inaba","submitted_at":"2025-06-26T16:56:43Z","abstract_excerpt":"Sparse autoencoders (SAEs) have become an important tool for analyzing and interpreting the activation space of transformer-based language models (LMs). However, SAEs suffer several shortcomings that diminish their utility and internal validity. Since SAEs are trained post-hoc, it is unclear if the failure to discover a particular concept is a failure on the SAE's side or due to the underlying LM not representing this concept. This problem is exacerbated by training conditions and architecture choices affecting which features an SAE learns. When tracing how LMs learn concepts during training, "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.21468","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/2506.21468/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":"2506.21468","created_at":"2026-07-05T11:27:46.470297+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.21468v1","created_at":"2026-07-05T11:27:46.470297+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.21468","created_at":"2026-07-05T11:27:46.470297+00:00"},{"alias_kind":"pith_short_12","alias_value":"UPRZD74KTMQL","created_at":"2026-07-05T11:27:46.470297+00:00"},{"alias_kind":"pith_short_16","alias_value":"UPRZD74KTMQLKHXE","created_at":"2026-07-05T11:27:46.470297+00:00"},{"alias_kind":"pith_short_8","alias_value":"UPRZD74K","created_at":"2026-07-05T11:27:46.470297+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/UPRZD74KTMQLKHXE7SFDIYXDAN","json":"https://pith.science/pith/UPRZD74KTMQLKHXE7SFDIYXDAN.json","graph_json":"https://pith.science/api/pith-number/UPRZD74KTMQLKHXE7SFDIYXDAN/graph.json","events_json":"https://pith.science/api/pith-number/UPRZD74KTMQLKHXE7SFDIYXDAN/events.json","paper":"https://pith.science/paper/UPRZD74K"},"agent_actions":{"view_html":"https://pith.science/pith/UPRZD74KTMQLKHXE7SFDIYXDAN","download_json":"https://pith.science/pith/UPRZD74KTMQLKHXE7SFDIYXDAN.json","view_paper":"https://pith.science/paper/UPRZD74K","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.21468&json=true","fetch_graph":"https://pith.science/api/pith-number/UPRZD74KTMQLKHXE7SFDIYXDAN/graph.json","fetch_events":"https://pith.science/api/pith-number/UPRZD74KTMQLKHXE7SFDIYXDAN/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/UPRZD74KTMQLKHXE7SFDIYXDAN/action/timestamp_anchor","attest_storage":"https://pith.science/pith/UPRZD74KTMQLKHXE7SFDIYXDAN/action/storage_attestation","attest_author":"https://pith.science/pith/UPRZD74KTMQLKHXE7SFDIYXDAN/action/author_attestation","sign_citation":"https://pith.science/pith/UPRZD74KTMQLKHXE7SFDIYXDAN/action/citation_signature","submit_replication":"https://pith.science/pith/UPRZD74KTMQLKHXE7SFDIYXDAN/action/replication_record"}},"created_at":"2026-07-05T11:27:46.470297+00:00","updated_at":"2026-07-05T11:27:46.470297+00:00"}