{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:JQRO5QHTNDZXWCWZCCFV6V2VWR","short_pith_number":"pith:JQRO5QHT","schema_version":"1.0","canonical_sha256":"4c22eec0f368f37b0ad9108b5f5755b46de52bf7ba86941fe6fb37a278ed254a","source":{"kind":"arxiv","id":"2512.22471","version":5},"attestation_state":"computed","paper":{"title":"The Bayesian Geometry of Transformer Attention","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","stat.ML"],"primary_cat":"cs.LG","authors_text":"Naman Agarwal, Siddhartha R. Dalal, Vishal Misra","submitted_at":"2025-12-27T05:28:58Z","abstract_excerpt":"Transformers often appear to perform Bayesian reasoning in context, but verifying this rigorously has been impossible: natural data lack analytic posteriors, and large models conflate reasoning with memorization. We address this by constructing \\emph{Bayesian wind tunnels} -- controlled environments where the true posterior is known in closed form and memorization is provably impossible. In these settings, small transformers reproduce Bayesian posteriors with $10^{-3}$-$10^{-4}$ bit accuracy, while capacity-matched MLPs fail by orders of magnitude, establishing a clear architectural separation"},"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":"2512.22471","kind":"arxiv","version":5},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-12-27T05:28:58Z","cross_cats_sorted":["cs.AI","stat.ML"],"title_canon_sha256":"6d5ac9abff865e48661c8fea616899b3472113d929ef4b02fa4f1859dddf6aa8","abstract_canon_sha256":"79396ae356a856532671a89c1f6c12436818037fb4fb06d310c12c430b6e6d86"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T00:03:00.885103Z","signature_b64":"zv0niTaDfLcenVZJd+PYD8rmLx3faCaB7pipGruGdSpfJIbRGa96jPoudLgzF7BEAVIKA4I4yzGNtzTZZdiPCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"4c22eec0f368f37b0ad9108b5f5755b46de52bf7ba86941fe6fb37a278ed254a","last_reissued_at":"2026-05-20T00:03:00.884206Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T00:03:00.884206Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"The Bayesian Geometry of Transformer Attention","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","stat.ML"],"primary_cat":"cs.LG","authors_text":"Naman Agarwal, Siddhartha R. Dalal, Vishal Misra","submitted_at":"2025-12-27T05:28:58Z","abstract_excerpt":"Transformers often appear to perform Bayesian reasoning in context, but verifying this rigorously has been impossible: natural data lack analytic posteriors, and large models conflate reasoning with memorization. We address this by constructing \\emph{Bayesian wind tunnels} -- controlled environments where the true posterior is known in closed form and memorization is provably impossible. In these settings, small transformers reproduce Bayesian posteriors with $10^{-3}$-$10^{-4}$ bit accuracy, while capacity-matched MLPs fail by orders of magnitude, establishing a clear architectural separation"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2512.22471","kind":"arxiv","version":5},"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/2512.22471/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":"2512.22471","created_at":"2026-05-20T00:03:00.884367+00:00"},{"alias_kind":"arxiv_version","alias_value":"2512.22471v5","created_at":"2026-05-20T00:03:00.884367+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2512.22471","created_at":"2026-05-20T00:03:00.884367+00:00"},{"alias_kind":"pith_short_12","alias_value":"JQRO5QHTNDZX","created_at":"2026-05-20T00:03:00.884367+00:00"},{"alias_kind":"pith_short_16","alias_value":"JQRO5QHTNDZXWCWZ","created_at":"2026-05-20T00:03:00.884367+00:00"},{"alias_kind":"pith_short_8","alias_value":"JQRO5QHT","created_at":"2026-05-20T00:03:00.884367+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":7,"internal_anchor_count":7,"sample":[{"citing_arxiv_id":"2512.22473","citing_title":"Gradient Dynamics of Attention: How Cross-Entropy Sculpts Bayesian Manifolds","ref_index":1,"is_internal_anchor":true},{"citing_arxiv_id":"2605.16665","citing_title":"In-context learning enables continental-scale subsurface temperature prediction from sparse local observations","ref_index":33,"is_internal_anchor":true},{"citing_arxiv_id":"2601.18832","citing_title":"The Geometric Reasoner: Manifold-Informed Latent Foresight Search for Long-Context Reasoning","ref_index":2,"is_internal_anchor":true},{"citing_arxiv_id":"2605.05686","citing_title":"Attractor Geometry of Transformer Memory: From Conflict Arbitration to Confident Hallucination","ref_index":1,"is_internal_anchor":true},{"citing_arxiv_id":"2605.13219","citing_title":"Comparative analysis of missing data imputation methods for CSST survey: Impact on photometric redshift estimation performance","ref_index":2,"is_internal_anchor":true},{"citing_arxiv_id":"2605.05686","citing_title":"Attractor Geometry of Transformer Memory: From Conflict Arbitration to Confident Hallucination","ref_index":1,"is_internal_anchor":true},{"citing_arxiv_id":"2605.00742","citing_title":"Position: agentic AI orchestration should be Bayes-consistent","ref_index":1,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/JQRO5QHTNDZXWCWZCCFV6V2VWR","json":"https://pith.science/pith/JQRO5QHTNDZXWCWZCCFV6V2VWR.json","graph_json":"https://pith.science/api/pith-number/JQRO5QHTNDZXWCWZCCFV6V2VWR/graph.json","events_json":"https://pith.science/api/pith-number/JQRO5QHTNDZXWCWZCCFV6V2VWR/events.json","paper":"https://pith.science/paper/JQRO5QHT"},"agent_actions":{"view_html":"https://pith.science/pith/JQRO5QHTNDZXWCWZCCFV6V2VWR","download_json":"https://pith.science/pith/JQRO5QHTNDZXWCWZCCFV6V2VWR.json","view_paper":"https://pith.science/paper/JQRO5QHT","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2512.22471&json=true","fetch_graph":"https://pith.science/api/pith-number/JQRO5QHTNDZXWCWZCCFV6V2VWR/graph.json","fetch_events":"https://pith.science/api/pith-number/JQRO5QHTNDZXWCWZCCFV6V2VWR/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/JQRO5QHTNDZXWCWZCCFV6V2VWR/action/timestamp_anchor","attest_storage":"https://pith.science/pith/JQRO5QHTNDZXWCWZCCFV6V2VWR/action/storage_attestation","attest_author":"https://pith.science/pith/JQRO5QHTNDZXWCWZCCFV6V2VWR/action/author_attestation","sign_citation":"https://pith.science/pith/JQRO5QHTNDZXWCWZCCFV6V2VWR/action/citation_signature","submit_replication":"https://pith.science/pith/JQRO5QHTNDZXWCWZCCFV6V2VWR/action/replication_record"}},"created_at":"2026-05-20T00:03:00.884367+00:00","updated_at":"2026-05-20T00:03:00.884367+00:00"}