{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:LFXQQWZA22B7BBUX3VPVGXFUBL","short_pith_number":"pith:LFXQQWZA","schema_version":"1.0","canonical_sha256":"596f085b20d683f08697dd5f535cb40ad0f98edba8c33dc19a43a876db167809","source":{"kind":"arxiv","id":"2306.15091","version":1},"attestation_state":"computed","paper":{"title":"Understanding In-Context Learning via Supportive Pretraining Data","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Asli Celikyilmaz, Daniel Simig, Tianlu Wang, Todor Mihaylov, Xiaochuang Han, Yulia Tsvetkov","submitted_at":"2023-06-26T22:14:04Z","abstract_excerpt":"In-context learning (ICL) improves language models' performance on a variety of NLP tasks by simply demonstrating a handful of examples at inference time. It is not well understood why ICL ability emerges, as the model has never been specifically trained on such demonstrations. Unlike prior work that explores implicit mechanisms behind ICL, we study ICL via investigating the pretraining data. Specifically, we first adapt an iterative, gradient-based approach to find a small subset of pretraining data that supports ICL. We observe that a continued pretraining on this small subset significantly "},"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":"2306.15091","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2023-06-26T22:14:04Z","cross_cats_sorted":[],"title_canon_sha256":"7c095914ff46b49a428cd3a8565a2cab344eb14178a09cd3909945ff603e523e","abstract_canon_sha256":"9b05cc338f912f66014a928f5c74cb4813c94670b7b8be15f124f7bdbca94390"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T06:24:53.690237Z","signature_b64":"ZxUsiEDWe8vNi7KgWzA6PLPclyyaw1xWg7L1SSwd5QIpTVGhjFY36k7KrrDDJn5TllXFxWNgamH3m/KkctQxDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"596f085b20d683f08697dd5f535cb40ad0f98edba8c33dc19a43a876db167809","last_reissued_at":"2026-07-05T06:24:53.689741Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T06:24:53.689741Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Understanding In-Context Learning via Supportive Pretraining Data","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Asli Celikyilmaz, Daniel Simig, Tianlu Wang, Todor Mihaylov, Xiaochuang Han, Yulia Tsvetkov","submitted_at":"2023-06-26T22:14:04Z","abstract_excerpt":"In-context learning (ICL) improves language models' performance on a variety of NLP tasks by simply demonstrating a handful of examples at inference time. It is not well understood why ICL ability emerges, as the model has never been specifically trained on such demonstrations. Unlike prior work that explores implicit mechanisms behind ICL, we study ICL via investigating the pretraining data. Specifically, we first adapt an iterative, gradient-based approach to find a small subset of pretraining data that supports ICL. We observe that a continued pretraining on this small subset significantly "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2306.15091","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/2306.15091/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":"2306.15091","created_at":"2026-07-05T06:24:53.689802+00:00"},{"alias_kind":"arxiv_version","alias_value":"2306.15091v1","created_at":"2026-07-05T06:24:53.689802+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2306.15091","created_at":"2026-07-05T06:24:53.689802+00:00"},{"alias_kind":"pith_short_12","alias_value":"LFXQQWZA22B7","created_at":"2026-07-05T06:24:53.689802+00:00"},{"alias_kind":"pith_short_16","alias_value":"LFXQQWZA22B7BBUX","created_at":"2026-07-05T06:24:53.689802+00:00"},{"alias_kind":"pith_short_8","alias_value":"LFXQQWZA","created_at":"2026-07-05T06:24:53.689802+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2604.13403","citing_title":"Why Multimodal In-Context Learning Lags Behind? Unveiling the Inner Mechanisms and Bottlenecks","ref_index":14,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/LFXQQWZA22B7BBUX3VPVGXFUBL","json":"https://pith.science/pith/LFXQQWZA22B7BBUX3VPVGXFUBL.json","graph_json":"https://pith.science/api/pith-number/LFXQQWZA22B7BBUX3VPVGXFUBL/graph.json","events_json":"https://pith.science/api/pith-number/LFXQQWZA22B7BBUX3VPVGXFUBL/events.json","paper":"https://pith.science/paper/LFXQQWZA"},"agent_actions":{"view_html":"https://pith.science/pith/LFXQQWZA22B7BBUX3VPVGXFUBL","download_json":"https://pith.science/pith/LFXQQWZA22B7BBUX3VPVGXFUBL.json","view_paper":"https://pith.science/paper/LFXQQWZA","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2306.15091&json=true","fetch_graph":"https://pith.science/api/pith-number/LFXQQWZA22B7BBUX3VPVGXFUBL/graph.json","fetch_events":"https://pith.science/api/pith-number/LFXQQWZA22B7BBUX3VPVGXFUBL/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/LFXQQWZA22B7BBUX3VPVGXFUBL/action/timestamp_anchor","attest_storage":"https://pith.science/pith/LFXQQWZA22B7BBUX3VPVGXFUBL/action/storage_attestation","attest_author":"https://pith.science/pith/LFXQQWZA22B7BBUX3VPVGXFUBL/action/author_attestation","sign_citation":"https://pith.science/pith/LFXQQWZA22B7BBUX3VPVGXFUBL/action/citation_signature","submit_replication":"https://pith.science/pith/LFXQQWZA22B7BBUX3VPVGXFUBL/action/replication_record"}},"created_at":"2026-07-05T06:24:53.689802+00:00","updated_at":"2026-07-05T06:24:53.689802+00:00"}