{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:W7HRHUDFOXCKQGYOMEUIBBS2MB","short_pith_number":"pith:W7HRHUDF","schema_version":"1.0","canonical_sha256":"b7cf13d06575c4a81b0e612880865a6075972f91d888b692ab7efde452109fed","source":{"kind":"arxiv","id":"2209.15189","version":1},"attestation_state":"computed","paper":{"title":"Learning by Distilling Context","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Charlie Snell, Dan Klein, Ruiqi Zhong","submitted_at":"2022-09-30T02:30:15Z","abstract_excerpt":"Language models significantly benefit from context tokens, such as prompts or scratchpads. They perform better when prompted with informative instructions, and they acquire new reasoning capabilities by generating a scratch-pad before predicting the final answers. However, they do not \\textit{internalize} these performance gains, which disappear when the context tokens are gone. Our work proposes to apply context distillation so that a language model can improve itself by internalizing these gains. Concretely, given a synthetic unlabeled input for the target task, we condition the model on ``["},"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":"2209.15189","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2022-09-30T02:30:15Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"68ca7d76412b1510d49f6cfc84058a5ec8b444203116e661455a2ae714aef384","abstract_canon_sha256":"d42e2b5e2e0712636fa29ed19b717aa72d013221caf6a0ae794943f7dcc34e39"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T05:02:17.253111Z","signature_b64":"Es7jAjdsxVt4o/M127q9gu3ZclQO8ipAL+S1cfiA1RVX85ym5pDDNIGGyUPYYMb8FzbA/Z2FM4pST3mS9ZUUDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"b7cf13d06575c4a81b0e612880865a6075972f91d888b692ab7efde452109fed","last_reissued_at":"2026-07-05T05:02:17.252719Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T05:02:17.252719Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Learning by Distilling Context","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Charlie Snell, Dan Klein, Ruiqi Zhong","submitted_at":"2022-09-30T02:30:15Z","abstract_excerpt":"Language models significantly benefit from context tokens, such as prompts or scratchpads. They perform better when prompted with informative instructions, and they acquire new reasoning capabilities by generating a scratch-pad before predicting the final answers. However, they do not \\textit{internalize} these performance gains, which disappear when the context tokens are gone. Our work proposes to apply context distillation so that a language model can improve itself by internalizing these gains. Concretely, given a synthetic unlabeled input for the target task, we condition the model on ``["},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2209.15189","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/2209.15189/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":"2209.15189","created_at":"2026-07-05T05:02:17.252775+00:00"},{"alias_kind":"arxiv_version","alias_value":"2209.15189v1","created_at":"2026-07-05T05:02:17.252775+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2209.15189","created_at":"2026-07-05T05:02:17.252775+00:00"},{"alias_kind":"pith_short_12","alias_value":"W7HRHUDFOXCK","created_at":"2026-07-05T05:02:17.252775+00:00"},{"alias_kind":"pith_short_16","alias_value":"W7HRHUDFOXCKQGYO","created_at":"2026-07-05T05:02:17.252775+00:00"},{"alias_kind":"pith_short_8","alias_value":"W7HRHUDF","created_at":"2026-07-05T05:02:17.252775+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":28,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.08032","citing_title":"What to Keep, What to Forget: A Rate--Distortion View of Memory Compaction in LLMs and Agents","ref_index":107,"is_internal_anchor":true},{"citing_arxiv_id":"2606.12942","citing_title":"PRISMR: Overcoming Parse Collapse in Multimodal Listwise Ranking via Parameterized Representation Internalization","ref_index":28,"is_internal_anchor":false},{"citing_arxiv_id":"2607.01690","citing_title":"Epistemic Goggles: A Pretrained Module that Induces an Epistemic Frame via Gradient Editing","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2606.12400","citing_title":"Doc-to-Atom: Learning to Compile and Compose Memory Atoms","ref_index":17,"is_internal_anchor":false},{"citing_arxiv_id":"2606.11853","citing_title":"Task-Aware Structured Memory for Dynamic Multi-modal In-Context Learning","ref_index":205,"is_internal_anchor":false},{"citing_arxiv_id":"2606.06154","citing_title":"Amortizing Federated Adaptation: Hypernetwork Driven LoRA for Personalized Foundation Models","ref_index":76,"is_internal_anchor":false},{"citing_arxiv_id":"2607.01208","citing_title":"Distill to Detect: Exposing Stealth Biases in LLMs through Cartridge Distillation","ref_index":28,"is_internal_anchor":false},{"citing_arxiv_id":"2606.04703","citing_title":"Rethinking Continual Experience Internalization for Self-Evolving LLM Agents","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2606.30775","citing_title":"A Single Rewrite Suffices: Empirical Lessons from Production Skill Description Optimization","ref_index":69,"is_internal_anchor":false},{"citing_arxiv_id":"2606.28349","citing_title":"HMARS: A Hierarchical Multi-Agent Memory System for Long-Context Reasoning","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2605.26099","citing_title":"Do Language Models Need Sleep? Offline Recurrence for Improved Online Inference","ref_index":52,"is_internal_anchor":false},{"citing_arxiv_id":"2606.01080","citing_title":"ThinkSwitch: Context Distillation with LoRA and Weight Interpolation for Specific-Purpose Reasoning Tasks","ref_index":23,"is_internal_anchor":false},{"citing_arxiv_id":"2605.22263","citing_title":"Tailoring Teaching to Aptitude: Direction-Adaptive Self-Distillation for LLM Reasoning","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2605.20258","citing_title":"It Takes Two: Complementary Self-Distillation for Contextual Integrity in LLMs","ref_index":39,"is_internal_anchor":false},{"citing_arxiv_id":"2605.15604","citing_title":"VSPO: Vector-Steered Policy Optimization for Behavioral Control","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2605.17497","citing_title":"Self-Supervised On-Policy Distillation for Reasoning Language Models","ref_index":1,"is_internal_anchor":false},{"citing_arxiv_id":"2605.18226","citing_title":"Context Memorization for Efficient Long Context Generation","ref_index":36,"is_internal_anchor":false},{"citing_arxiv_id":"2210.11610","citing_title":"Large Language Models Can Self-Improve","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2406.10162","citing_title":"Sycophancy to Subterfuge: Investigating Reward-Tampering in Large Language Models","ref_index":297,"is_internal_anchor":false},{"citing_arxiv_id":"2604.09571","citing_title":"Tuning Qwen2.5-VL to Improve Its Web Interaction Skills","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2605.11613","citing_title":"From Generic Correlation to Input-Specific Credit in On-Policy Self Distillation","ref_index":33,"is_internal_anchor":false},{"citing_arxiv_id":"2404.13208","citing_title":"The Instruction Hierarchy: Training LLMs to Prioritize Privileged Instructions","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08873","citing_title":"CoDistill-GRPO: A Co-Distillation Recipe for Efficient Group Relative Policy Optimization","ref_index":28,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10889","citing_title":"Unmasking On-Policy Distillation: Where It Helps, Where It Hurts, and Why","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2604.07894","citing_title":"TSUBASA: Improving Long-Horizon Personalization via Evolving Memory and Self-Learning with Context Distillation","ref_index":58,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/W7HRHUDFOXCKQGYOMEUIBBS2MB","json":"https://pith.science/pith/W7HRHUDFOXCKQGYOMEUIBBS2MB.json","graph_json":"https://pith.science/api/pith-number/W7HRHUDFOXCKQGYOMEUIBBS2MB/graph.json","events_json":"https://pith.science/api/pith-number/W7HRHUDFOXCKQGYOMEUIBBS2MB/events.json","paper":"https://pith.science/paper/W7HRHUDF"},"agent_actions":{"view_html":"https://pith.science/pith/W7HRHUDFOXCKQGYOMEUIBBS2MB","download_json":"https://pith.science/pith/W7HRHUDFOXCKQGYOMEUIBBS2MB.json","view_paper":"https://pith.science/paper/W7HRHUDF","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2209.15189&json=true","fetch_graph":"https://pith.science/api/pith-number/W7HRHUDFOXCKQGYOMEUIBBS2MB/graph.json","fetch_events":"https://pith.science/api/pith-number/W7HRHUDFOXCKQGYOMEUIBBS2MB/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/W7HRHUDFOXCKQGYOMEUIBBS2MB/action/timestamp_anchor","attest_storage":"https://pith.science/pith/W7HRHUDFOXCKQGYOMEUIBBS2MB/action/storage_attestation","attest_author":"https://pith.science/pith/W7HRHUDFOXCKQGYOMEUIBBS2MB/action/author_attestation","sign_citation":"https://pith.science/pith/W7HRHUDFOXCKQGYOMEUIBBS2MB/action/citation_signature","submit_replication":"https://pith.science/pith/W7HRHUDFOXCKQGYOMEUIBBS2MB/action/replication_record"}},"created_at":"2026-07-05T05:02:17.252775+00:00","updated_at":"2026-07-05T05:02:17.252775+00:00"}