{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:MLCP46YTSB74PVO6YN7RZ725HY","short_pith_number":"pith:MLCP46YT","schema_version":"1.0","canonical_sha256":"62c4fe7b13907fc7d5dec37f1cff5d3e187bb875e1a1315e1f37e5f72662336d","source":{"kind":"arxiv","id":"2401.17505","version":4},"attestation_state":"computed","paper":{"title":"Arrows of Time for Large Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Cl\\'ement Hongler, J\\'er\\'emie Wenger, Vassilis Papadopoulos","submitted_at":"2024-01-30T23:46:35Z","abstract_excerpt":"We study the probabilistic modeling performed by Autoregressive Large Language Models (LLMs) through the angle of time directionality, addressing a question first raised in (Shannon, 1951). For large enough models, we empirically find a time asymmetry in their ability to learn natural language: a difference in the average log-perplexity when trying to predict the next token versus when trying to predict the previous one. This difference is at the same time subtle and very consistent across various modalities (language, model size, training time, ...). Theoretically, this is surprising: from 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":"2401.17505","kind":"arxiv","version":4},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-01-30T23:46:35Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"1bbe59d146f27eb0d2de23d178dcc95bcac7a2dd478773010ab877c2b0e3a8cc","abstract_canon_sha256":"58aebbc54fe6887fb7bba4f21641e83984c0ddbc2ce085ac692c0df0d9aa2d0d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-29T01:14:00.122302Z","signature_b64":"hCV0SHgt/vXYj6m3mloARDpsaU7uJntAaDEszznZ58SnaCZfamPsAyhTrHcWG9I673KtAzjxdYU4LYtkllZaCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"62c4fe7b13907fc7d5dec37f1cff5d3e187bb875e1a1315e1f37e5f72662336d","last_reissued_at":"2026-06-29T01:14:00.121786Z","signature_status":"signed_v1","first_computed_at":"2026-06-29T01:14:00.121786Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Arrows of Time for Large Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Cl\\'ement Hongler, J\\'er\\'emie Wenger, Vassilis Papadopoulos","submitted_at":"2024-01-30T23:46:35Z","abstract_excerpt":"We study the probabilistic modeling performed by Autoregressive Large Language Models (LLMs) through the angle of time directionality, addressing a question first raised in (Shannon, 1951). For large enough models, we empirically find a time asymmetry in their ability to learn natural language: a difference in the average log-perplexity when trying to predict the next token versus when trying to predict the previous one. This difference is at the same time subtle and very consistent across various modalities (language, model size, training time, ...). Theoretically, this is surprising: from an"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2401.17505","kind":"arxiv","version":4},"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/2401.17505/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":"2401.17505","created_at":"2026-06-29T01:14:00.121847+00:00"},{"alias_kind":"arxiv_version","alias_value":"2401.17505v4","created_at":"2026-06-29T01:14:00.121847+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2401.17505","created_at":"2026-06-29T01:14:00.121847+00:00"},{"alias_kind":"pith_short_12","alias_value":"MLCP46YTSB74","created_at":"2026-06-29T01:14:00.121847+00:00"},{"alias_kind":"pith_short_16","alias_value":"MLCP46YTSB74PVO6","created_at":"2026-06-29T01:14:00.121847+00:00"},{"alias_kind":"pith_short_8","alias_value":"MLCP46YT","created_at":"2026-06-29T01:14:00.121847+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2504.17069","citing_title":"Distilling Specialized Orders for Visual Generation","ref_index":8,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/MLCP46YTSB74PVO6YN7RZ725HY","json":"https://pith.science/pith/MLCP46YTSB74PVO6YN7RZ725HY.json","graph_json":"https://pith.science/api/pith-number/MLCP46YTSB74PVO6YN7RZ725HY/graph.json","events_json":"https://pith.science/api/pith-number/MLCP46YTSB74PVO6YN7RZ725HY/events.json","paper":"https://pith.science/paper/MLCP46YT"},"agent_actions":{"view_html":"https://pith.science/pith/MLCP46YTSB74PVO6YN7RZ725HY","download_json":"https://pith.science/pith/MLCP46YTSB74PVO6YN7RZ725HY.json","view_paper":"https://pith.science/paper/MLCP46YT","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2401.17505&json=true","fetch_graph":"https://pith.science/api/pith-number/MLCP46YTSB74PVO6YN7RZ725HY/graph.json","fetch_events":"https://pith.science/api/pith-number/MLCP46YTSB74PVO6YN7RZ725HY/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/MLCP46YTSB74PVO6YN7RZ725HY/action/timestamp_anchor","attest_storage":"https://pith.science/pith/MLCP46YTSB74PVO6YN7RZ725HY/action/storage_attestation","attest_author":"https://pith.science/pith/MLCP46YTSB74PVO6YN7RZ725HY/action/author_attestation","sign_citation":"https://pith.science/pith/MLCP46YTSB74PVO6YN7RZ725HY/action/citation_signature","submit_replication":"https://pith.science/pith/MLCP46YTSB74PVO6YN7RZ725HY/action/replication_record"}},"created_at":"2026-06-29T01:14:00.121847+00:00","updated_at":"2026-06-29T01:14:00.121847+00:00"}