{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:2R326XCGD57XWBYBFTAKVUSSBL","short_pith_number":"pith:2R326XCG","schema_version":"1.0","canonical_sha256":"d477af5c461f7f7b07012cc0aad2520acee46d2fcbeb780ab61b0d1140916dfe","source":{"kind":"arxiv","id":"2606.26749","version":1},"attestation_state":"computed","paper":{"title":"Structure Before Collapse: Transient semantic geometry in next-token prediction","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Christos Thrampoulidis, Isabel Papadimitriou, Yize Zhao","submitted_at":"2026-06-25T08:33:34Z","abstract_excerpt":"Neural Collapse predicts that balanced one-hot classification pushes model representations to be equally far from each other; a symmetric configuration that depends only on the output label and ignores any semantic similarity in the inputs. This creates a puzzle: next-token prediction language models are trained predominantly (as context length increases) with one-hot labels: the same context is very unlikely to appear twice in training with different labels. However, they clearly learn latent structural features. That is, despite the one-hot training regime, a language model's contextual embe"},"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":"2606.26749","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-06-25T08:33:34Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"4d9a238edcaf3093781ff37aef48398d9d3c0ca77d4ca90b680c283fe8f560fd","abstract_canon_sha256":"1730a138a0e44eaf6984cacb50266856b2962d5e676eab05c328e2bc76776fee"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-26T01:15:58.695795Z","signature_b64":"a93mF3rueZh0ORUPFO5EARwoE+MOsCANpUt5rqiURc9EXjLtrnb79av2zL+qn4uD6tJhRGwVW7w9Ma7kV62vCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"d477af5c461f7f7b07012cc0aad2520acee46d2fcbeb780ab61b0d1140916dfe","last_reissued_at":"2026-06-26T01:15:58.695385Z","signature_status":"signed_v1","first_computed_at":"2026-06-26T01:15:58.695385Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Structure Before Collapse: Transient semantic geometry in next-token prediction","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Christos Thrampoulidis, Isabel Papadimitriou, Yize Zhao","submitted_at":"2026-06-25T08:33:34Z","abstract_excerpt":"Neural Collapse predicts that balanced one-hot classification pushes model representations to be equally far from each other; a symmetric configuration that depends only on the output label and ignores any semantic similarity in the inputs. This creates a puzzle: next-token prediction language models are trained predominantly (as context length increases) with one-hot labels: the same context is very unlikely to appear twice in training with different labels. However, they clearly learn latent structural features. That is, despite the one-hot training regime, a language model's contextual embe"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.26749","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/2606.26749/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":"2606.26749","created_at":"2026-06-26T01:15:58.695442+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.26749v1","created_at":"2026-06-26T01:15:58.695442+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.26749","created_at":"2026-06-26T01:15:58.695442+00:00"},{"alias_kind":"pith_short_12","alias_value":"2R326XCGD57X","created_at":"2026-06-26T01:15:58.695442+00:00"},{"alias_kind":"pith_short_16","alias_value":"2R326XCGD57XWBYB","created_at":"2026-06-26T01:15:58.695442+00:00"},{"alias_kind":"pith_short_8","alias_value":"2R326XCG","created_at":"2026-06-26T01:15:58.695442+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/2R326XCGD57XWBYBFTAKVUSSBL","json":"https://pith.science/pith/2R326XCGD57XWBYBFTAKVUSSBL.json","graph_json":"https://pith.science/api/pith-number/2R326XCGD57XWBYBFTAKVUSSBL/graph.json","events_json":"https://pith.science/api/pith-number/2R326XCGD57XWBYBFTAKVUSSBL/events.json","paper":"https://pith.science/paper/2R326XCG"},"agent_actions":{"view_html":"https://pith.science/pith/2R326XCGD57XWBYBFTAKVUSSBL","download_json":"https://pith.science/pith/2R326XCGD57XWBYBFTAKVUSSBL.json","view_paper":"https://pith.science/paper/2R326XCG","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.26749&json=true","fetch_graph":"https://pith.science/api/pith-number/2R326XCGD57XWBYBFTAKVUSSBL/graph.json","fetch_events":"https://pith.science/api/pith-number/2R326XCGD57XWBYBFTAKVUSSBL/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/2R326XCGD57XWBYBFTAKVUSSBL/action/timestamp_anchor","attest_storage":"https://pith.science/pith/2R326XCGD57XWBYBFTAKVUSSBL/action/storage_attestation","attest_author":"https://pith.science/pith/2R326XCGD57XWBYBFTAKVUSSBL/action/author_attestation","sign_citation":"https://pith.science/pith/2R326XCGD57XWBYBFTAKVUSSBL/action/citation_signature","submit_replication":"https://pith.science/pith/2R326XCGD57XWBYBFTAKVUSSBL/action/replication_record"}},"created_at":"2026-06-26T01:15:58.695442+00:00","updated_at":"2026-06-26T01:15:58.695442+00:00"}