{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:CPVBP4VGSBTOPVUZKLXO7GONFN","short_pith_number":"pith:CPVBP4VG","schema_version":"1.0","canonical_sha256":"13ea17f2a69066e7d69952eeef99cd2b6bdf95fae2657e0adb998a12c323d68d","source":{"kind":"arxiv","id":"2502.05253","version":1},"attestation_state":"computed","paper":{"title":"LLMs Can Teach Themselves to Better Predict the Future","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Benjamin Turtel, Danny Franklin, Philipp Schoenegger","submitted_at":"2025-02-07T17:21:16Z","abstract_excerpt":"We present an outcome-driven fine-tuning framework that enhances the forecasting capabilities of large language models (LLMs) without relying on human-curated reasoning samples. Our method leverages model self-play to generate pairs of diverse reasoning trajectories and probabilistic forecasts for a set of diverse questions that resolve after the models' knowledge cutoff date. We then rank pairs of these reasoning traces by their distance to the actual outcomes before fine-tuning the model via Direct Preference Optimization (DPO). On a separate test set, our approach increases prediction accur"},"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":"2502.05253","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-02-07T17:21:16Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"e16e9b17d7db949b0b58994a792f951b8d28a51a410751ebbaedbe9212a28dae","abstract_canon_sha256":"4fdbdbe328864d431b0f6e1dfac84cf557915154f1dd9ce9658ad47434650bcb"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:11:34.949609Z","signature_b64":"UEr1tUQ35zVfbumOI90bdZ4tos/iJ/k17GX4ZGKnOyfoFzXGxL82DKwc8UhoatuhokUqjzFrHJ7cH2G7sy+8Dg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"13ea17f2a69066e7d69952eeef99cd2b6bdf95fae2657e0adb998a12c323d68d","last_reissued_at":"2026-07-05T10:11:34.949080Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:11:34.949080Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"LLMs Can Teach Themselves to Better Predict the Future","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Benjamin Turtel, Danny Franklin, Philipp Schoenegger","submitted_at":"2025-02-07T17:21:16Z","abstract_excerpt":"We present an outcome-driven fine-tuning framework that enhances the forecasting capabilities of large language models (LLMs) without relying on human-curated reasoning samples. Our method leverages model self-play to generate pairs of diverse reasoning trajectories and probabilistic forecasts for a set of diverse questions that resolve after the models' knowledge cutoff date. We then rank pairs of these reasoning traces by their distance to the actual outcomes before fine-tuning the model via Direct Preference Optimization (DPO). On a separate test set, our approach increases prediction accur"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.05253","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/2502.05253/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":"2502.05253","created_at":"2026-07-05T10:11:34.949143+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.05253v1","created_at":"2026-07-05T10:11:34.949143+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.05253","created_at":"2026-07-05T10:11:34.949143+00:00"},{"alias_kind":"pith_short_12","alias_value":"CPVBP4VGSBTO","created_at":"2026-07-05T10:11:34.949143+00:00"},{"alias_kind":"pith_short_16","alias_value":"CPVBP4VGSBTOPVUZ","created_at":"2026-07-05T10:11:34.949143+00:00"},{"alias_kind":"pith_short_8","alias_value":"CPVBP4VG","created_at":"2026-07-05T10:11:34.949143+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2607.01661","citing_title":"Diverse Evidence, Better Forecasts: Multi-Agent Deliberation Under Information Asymmetry","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2607.00164","citing_title":"Verifiable Rewards for Calibrated Probabilistic Forecasting","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2605.26074","citing_title":"StakeBench: Evaluating Language Understanding Grounded in Market Commitment","ref_index":25,"is_internal_anchor":false},{"citing_arxiv_id":"2605.18768","citing_title":"ClinQueryAgent: A Conversational Agent for Population Health Management","ref_index":177,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/CPVBP4VGSBTOPVUZKLXO7GONFN","json":"https://pith.science/pith/CPVBP4VGSBTOPVUZKLXO7GONFN.json","graph_json":"https://pith.science/api/pith-number/CPVBP4VGSBTOPVUZKLXO7GONFN/graph.json","events_json":"https://pith.science/api/pith-number/CPVBP4VGSBTOPVUZKLXO7GONFN/events.json","paper":"https://pith.science/paper/CPVBP4VG"},"agent_actions":{"view_html":"https://pith.science/pith/CPVBP4VGSBTOPVUZKLXO7GONFN","download_json":"https://pith.science/pith/CPVBP4VGSBTOPVUZKLXO7GONFN.json","view_paper":"https://pith.science/paper/CPVBP4VG","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.05253&json=true","fetch_graph":"https://pith.science/api/pith-number/CPVBP4VGSBTOPVUZKLXO7GONFN/graph.json","fetch_events":"https://pith.science/api/pith-number/CPVBP4VGSBTOPVUZKLXO7GONFN/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/CPVBP4VGSBTOPVUZKLXO7GONFN/action/timestamp_anchor","attest_storage":"https://pith.science/pith/CPVBP4VGSBTOPVUZKLXO7GONFN/action/storage_attestation","attest_author":"https://pith.science/pith/CPVBP4VGSBTOPVUZKLXO7GONFN/action/author_attestation","sign_citation":"https://pith.science/pith/CPVBP4VGSBTOPVUZKLXO7GONFN/action/citation_signature","submit_replication":"https://pith.science/pith/CPVBP4VGSBTOPVUZKLXO7GONFN/action/replication_record"}},"created_at":"2026-07-05T10:11:34.949143+00:00","updated_at":"2026-07-05T10:11:34.949143+00:00"}