{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:GWTNS7TCROZLSTHLMCGBU5XGE7","short_pith_number":"pith:GWTNS7TC","schema_version":"1.0","canonical_sha256":"35a6d97e628bb2b94ceb608c1a76e627c4fd9c1fc89f5b2ff0838c4b807279c7","source":{"kind":"arxiv","id":"2506.17859","version":2},"attestation_state":"computed","paper":{"title":"In-Context Learning Strategies Emerge Rationally","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Core Francisco Park, Daniel Wurgaft, Ekdeep Singh Lubana, Gautam Reddy, Hidenori Tanaka, Noah D. Goodman","submitted_at":"2025-06-21T23:49:08Z","abstract_excerpt":"Recent work analyzing in-context learning (ICL) has identified a broad set of strategies that describe model behavior in different experimental conditions. We aim to unify these findings by asking why a model learns these disparate strategies in the first place. Specifically, we start with the observation that when trained to learn a mixture of tasks, as is popular in the literature, the strategies learned by a model for performing ICL can be captured by a family of Bayesian predictors: a memorizing predictor, which assumes a discrete prior on the set of seen tasks, and a generalizing predicto"},"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":"2506.17859","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-06-21T23:49:08Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"63320f91b308b150fab7911c6464254706ed2b6f5ef69116b2aca7998f901fe5","abstract_canon_sha256":"6a3823ee4ea8ac64b177c2e97a65d69d7d689ca816c6cc29bf3e390cb2213c40"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:27:31.400060Z","signature_b64":"Pdoo8szgkI7w6zqsn5iqgd+kJG6xkRwlUfUvxTVs3eOy4RPkE9RCfwp2YPaS9J3/TjJMXx5vbRydROJRN1hxAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"35a6d97e628bb2b94ceb608c1a76e627c4fd9c1fc89f5b2ff0838c4b807279c7","last_reissued_at":"2026-07-05T11:27:31.399571Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:27:31.399571Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"In-Context Learning Strategies Emerge Rationally","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Core Francisco Park, Daniel Wurgaft, Ekdeep Singh Lubana, Gautam Reddy, Hidenori Tanaka, Noah D. Goodman","submitted_at":"2025-06-21T23:49:08Z","abstract_excerpt":"Recent work analyzing in-context learning (ICL) has identified a broad set of strategies that describe model behavior in different experimental conditions. We aim to unify these findings by asking why a model learns these disparate strategies in the first place. Specifically, we start with the observation that when trained to learn a mixture of tasks, as is popular in the literature, the strategies learned by a model for performing ICL can be captured by a family of Bayesian predictors: a memorizing predictor, which assumes a discrete prior on the set of seen tasks, and a generalizing predicto"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.17859","kind":"arxiv","version":2},"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/2506.17859/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":"2506.17859","created_at":"2026-07-05T11:27:31.399631+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.17859v2","created_at":"2026-07-05T11:27:31.399631+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.17859","created_at":"2026-07-05T11:27:31.399631+00:00"},{"alias_kind":"pith_short_12","alias_value":"GWTNS7TCROZL","created_at":"2026-07-05T11:27:31.399631+00:00"},{"alias_kind":"pith_short_16","alias_value":"GWTNS7TCROZLSTHL","created_at":"2026-07-05T11:27:31.399631+00:00"},{"alias_kind":"pith_short_8","alias_value":"GWTNS7TC","created_at":"2026-07-05T11:27:31.399631+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.18284","citing_title":"Breaking the Solver Bottleneck: Training Task Generators at the Learnable Frontier","ref_index":97,"is_internal_anchor":false},{"citing_arxiv_id":"2509.23108","citing_title":"Artificial Phantasia: Emergent Mental Imagery in Large Language Models","ref_index":83,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12412","citing_title":"Stories in Space: In-Context Learning Trajectories in Conceptual Belief Space","ref_index":110,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/GWTNS7TCROZLSTHLMCGBU5XGE7","json":"https://pith.science/pith/GWTNS7TCROZLSTHLMCGBU5XGE7.json","graph_json":"https://pith.science/api/pith-number/GWTNS7TCROZLSTHLMCGBU5XGE7/graph.json","events_json":"https://pith.science/api/pith-number/GWTNS7TCROZLSTHLMCGBU5XGE7/events.json","paper":"https://pith.science/paper/GWTNS7TC"},"agent_actions":{"view_html":"https://pith.science/pith/GWTNS7TCROZLSTHLMCGBU5XGE7","download_json":"https://pith.science/pith/GWTNS7TCROZLSTHLMCGBU5XGE7.json","view_paper":"https://pith.science/paper/GWTNS7TC","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.17859&json=true","fetch_graph":"https://pith.science/api/pith-number/GWTNS7TCROZLSTHLMCGBU5XGE7/graph.json","fetch_events":"https://pith.science/api/pith-number/GWTNS7TCROZLSTHLMCGBU5XGE7/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/GWTNS7TCROZLSTHLMCGBU5XGE7/action/timestamp_anchor","attest_storage":"https://pith.science/pith/GWTNS7TCROZLSTHLMCGBU5XGE7/action/storage_attestation","attest_author":"https://pith.science/pith/GWTNS7TCROZLSTHLMCGBU5XGE7/action/author_attestation","sign_citation":"https://pith.science/pith/GWTNS7TCROZLSTHLMCGBU5XGE7/action/citation_signature","submit_replication":"https://pith.science/pith/GWTNS7TCROZLSTHLMCGBU5XGE7/action/replication_record"}},"created_at":"2026-07-05T11:27:31.399631+00:00","updated_at":"2026-07-05T11:27:31.399631+00:00"}