{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:NDZ7CBXJAHGCN7E7FF5EFBBIRH","short_pith_number":"pith:NDZ7CBXJ","schema_version":"1.0","canonical_sha256":"68f3f106e901cc26fc9f297a42842889d3202cc6732653f5d91af4eb9295f3c5","source":{"kind":"arxiv","id":"2411.07279","version":2},"attestation_state":"computed","paper":{"title":"The Surprising Effectiveness of Test-Time Training for Few-Shot Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL","cs.LG"],"primary_cat":"cs.AI","authors_text":"Adam Zweiger, Ekin Aky\\\"urek, Han Guo, Jacob Andreas, Jyothish Pari, Linlu Qiu, Mehul Damani, Yoon Kim","submitted_at":"2024-11-11T18:59:45Z","abstract_excerpt":"Language models (LMs) have shown impressive performance on tasks within their training distribution, but often struggle with structurally novel tasks even when given a small number of in-context task examples. We investigate the effectiveness of test-time training (TTT) -- temporarily updating model parameters during inference using a loss derived from input data -- as a mechanism for improving LMs' reasoning and few-shot learning capabilities. On the Abstraction and Reasoning Corpus (ARC), performing TTT with in-context examples yields up to $6\\times$ higher accuracy compared to fine-tuned ba"},"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":"2411.07279","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AI","submitted_at":"2024-11-11T18:59:45Z","cross_cats_sorted":["cs.CL","cs.LG"],"title_canon_sha256":"b80ba4f207a4de0d1067416714bf4ed7a25c3f2b4d5a63c184c06f9680bcf30e","abstract_canon_sha256":"792dd9899d2dd045eb9b26a8f06003c5ac47fa2ba7968117f8ec2e433e6d3264"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:38:37.864541Z","signature_b64":"F7/UOuvuYa20tvIMIMHdY+UNeCajcheVdMnJ17n9q3Fjt2pf+7ri6JHdn5vHS/dK4QfaCp9lDI7qSMvTXAXgCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"68f3f106e901cc26fc9f297a42842889d3202cc6732653f5d91af4eb9295f3c5","last_reissued_at":"2026-07-05T10:38:37.864053Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:38:37.864053Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"The Surprising Effectiveness of Test-Time Training for Few-Shot Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL","cs.LG"],"primary_cat":"cs.AI","authors_text":"Adam Zweiger, Ekin Aky\\\"urek, Han Guo, Jacob Andreas, Jyothish Pari, Linlu Qiu, Mehul Damani, Yoon Kim","submitted_at":"2024-11-11T18:59:45Z","abstract_excerpt":"Language models (LMs) have shown impressive performance on tasks within their training distribution, but often struggle with structurally novel tasks even when given a small number of in-context task examples. We investigate the effectiveness of test-time training (TTT) -- temporarily updating model parameters during inference using a loss derived from input data -- as a mechanism for improving LMs' reasoning and few-shot learning capabilities. On the Abstraction and Reasoning Corpus (ARC), performing TTT with in-context examples yields up to $6\\times$ higher accuracy compared to fine-tuned ba"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2411.07279","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/2411.07279/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":"2411.07279","created_at":"2026-07-05T10:38:37.864108+00:00"},{"alias_kind":"arxiv_version","alias_value":"2411.07279v2","created_at":"2026-07-05T10:38:37.864108+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2411.07279","created_at":"2026-07-05T10:38:37.864108+00:00"},{"alias_kind":"pith_short_12","alias_value":"NDZ7CBXJAHGC","created_at":"2026-07-05T10:38:37.864108+00:00"},{"alias_kind":"pith_short_16","alias_value":"NDZ7CBXJAHGCN7E7","created_at":"2026-07-05T10:38:37.864108+00:00"},{"alias_kind":"pith_short_8","alias_value":"NDZ7CBXJ","created_at":"2026-07-05T10:38:37.864108+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":13,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.21228","citing_title":"Sakana Fugu Technical Report","ref_index":182,"is_internal_anchor":false},{"citing_arxiv_id":"2606.06906","citing_title":"EASE-TTT: Evidence-Aligned Selective Test-Time Training for Long-Context Question Answering","ref_index":37,"is_internal_anchor":false},{"citing_arxiv_id":"2606.01070","citing_title":"Test-Time Training for Zero-Resource Dense Retrieval Reranking","ref_index":24,"is_internal_anchor":false},{"citing_arxiv_id":"2605.30337","citing_title":"Efficient Test-Time Finetuning of LLMs via Convex Reconstruction and Gradient Caching","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2509.19349","citing_title":"ShinkaEvolve: Towards Open-Ended And Sample-Efficient Program Evolution","ref_index":148,"is_internal_anchor":false},{"citing_arxiv_id":"2601.16175","citing_title":"Learning to Discover at Test Time","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2510.01346","citing_title":"Aristotle: IMO-level Automated Theorem Proving","ref_index":1,"is_internal_anchor":false},{"citing_arxiv_id":"2605.11328","citing_title":"Epistemic Uncertainty for Test-Time Discovery","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2604.27850","citing_title":"Reasoning over Object Descriptions Improves Coreference Resolution in Task-Based Dialogue Systems","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10237","citing_title":"The Benefits of Temporal Correlations: SGD Learns k-Juntas from Random Walks Efficiently","ref_index":132,"is_internal_anchor":false},{"citing_arxiv_id":"2605.01154","citing_title":"Multi-Perspective Transformers in ARC-AGI-2 Challenge","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2604.18839","citing_title":"One Step Forward and K Steps Back: Better Reasoning with Denoising Recursion Models","ref_index":159,"is_internal_anchor":false},{"citing_arxiv_id":"2604.18572","citing_title":"Back into Plato's Cave: Examining Cross-modal Representational Convergence at Scale","ref_index":2,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/NDZ7CBXJAHGCN7E7FF5EFBBIRH","json":"https://pith.science/pith/NDZ7CBXJAHGCN7E7FF5EFBBIRH.json","graph_json":"https://pith.science/api/pith-number/NDZ7CBXJAHGCN7E7FF5EFBBIRH/graph.json","events_json":"https://pith.science/api/pith-number/NDZ7CBXJAHGCN7E7FF5EFBBIRH/events.json","paper":"https://pith.science/paper/NDZ7CBXJ"},"agent_actions":{"view_html":"https://pith.science/pith/NDZ7CBXJAHGCN7E7FF5EFBBIRH","download_json":"https://pith.science/pith/NDZ7CBXJAHGCN7E7FF5EFBBIRH.json","view_paper":"https://pith.science/paper/NDZ7CBXJ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2411.07279&json=true","fetch_graph":"https://pith.science/api/pith-number/NDZ7CBXJAHGCN7E7FF5EFBBIRH/graph.json","fetch_events":"https://pith.science/api/pith-number/NDZ7CBXJAHGCN7E7FF5EFBBIRH/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/NDZ7CBXJAHGCN7E7FF5EFBBIRH/action/timestamp_anchor","attest_storage":"https://pith.science/pith/NDZ7CBXJAHGCN7E7FF5EFBBIRH/action/storage_attestation","attest_author":"https://pith.science/pith/NDZ7CBXJAHGCN7E7FF5EFBBIRH/action/author_attestation","sign_citation":"https://pith.science/pith/NDZ7CBXJAHGCN7E7FF5EFBBIRH/action/citation_signature","submit_replication":"https://pith.science/pith/NDZ7CBXJAHGCN7E7FF5EFBBIRH/action/replication_record"}},"created_at":"2026-07-05T10:38:37.864108+00:00","updated_at":"2026-07-05T10:38:37.864108+00:00"}