{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:SH25GJLM4ZVTRFMTOJENWKADXD","short_pith_number":"pith:SH25GJLM","schema_version":"1.0","canonical_sha256":"91f5d3256ce66b3895937248db2803b8d91a0c11d59240d5aec47639a26f1946","source":{"kind":"arxiv","id":"2606.11045","version":1},"attestation_state":"computed","paper":{"title":"What Fits (Into Few Tokens) Doesn't Overfit: Compression and Generalization in ML Research Agents","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.AI","authors_text":"Aaron Roth, Martin Andres Bertran, Zhiwei Steven Wu","submitted_at":"2026-06-09T16:12:55Z","abstract_excerpt":"Reusing a held-out benchmark adaptively should, in principle, invite overfitting. Yet benchmark-driven machine learning (ML) has produced surprisingly little overfitting in practice. An attractive hypothesis is that successful ML strategies are highly compressible. We study this in the setting of LLM-driven research agents, where the hypothesis becomes directly testable via two complementary information bottlenecks. In \\emph{output compression}, an exploration agent adaptively searches for high-performance models using a validation set, and we test whether a fresh ``reproducer agent'' can repr"},"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":"2606.11045","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-06-09T16:12:55Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"9e77f32abd30c68fa4254f0d8b34a1a14dcc944a56598d317fcdf3cd5c93b4e7","abstract_canon_sha256":"b6271d9783f25835fed65781142fb08768d40036e1376b2ccc3998b7c3400652"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-10T01:11:06.659768Z","signature_b64":"wPt1zaVkGFl5zSCOQDfrhgI3A+cTpaMywtgFrgmK2vl8fAgT6+DT1D8MaTK5kOIzT8WXo8PqFj20gezkTjcPAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"91f5d3256ce66b3895937248db2803b8d91a0c11d59240d5aec47639a26f1946","last_reissued_at":"2026-06-10T01:11:06.658905Z","signature_status":"signed_v1","first_computed_at":"2026-06-10T01:11:06.658905Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"What Fits (Into Few Tokens) Doesn't Overfit: Compression and Generalization in ML Research Agents","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.AI","authors_text":"Aaron Roth, Martin Andres Bertran, Zhiwei Steven Wu","submitted_at":"2026-06-09T16:12:55Z","abstract_excerpt":"Reusing a held-out benchmark adaptively should, in principle, invite overfitting. Yet benchmark-driven machine learning (ML) has produced surprisingly little overfitting in practice. An attractive hypothesis is that successful ML strategies are highly compressible. We study this in the setting of LLM-driven research agents, where the hypothesis becomes directly testable via two complementary information bottlenecks. In \\emph{output compression}, an exploration agent adaptively searches for high-performance models using a validation set, and we test whether a fresh ``reproducer agent'' can repr"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.11045","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/2606.11045/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":"2606.11045","created_at":"2026-06-10T01:11:06.659047+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.11045v1","created_at":"2026-06-10T01:11:06.659047+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.11045","created_at":"2026-06-10T01:11:06.659047+00:00"},{"alias_kind":"pith_short_12","alias_value":"SH25GJLM4ZVT","created_at":"2026-06-10T01:11:06.659047+00:00"},{"alias_kind":"pith_short_16","alias_value":"SH25GJLM4ZVTRFMT","created_at":"2026-06-10T01:11:06.659047+00:00"},{"alias_kind":"pith_short_8","alias_value":"SH25GJLM","created_at":"2026-06-10T01:11:06.659047+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/SH25GJLM4ZVTRFMTOJENWKADXD","json":"https://pith.science/pith/SH25GJLM4ZVTRFMTOJENWKADXD.json","graph_json":"https://pith.science/api/pith-number/SH25GJLM4ZVTRFMTOJENWKADXD/graph.json","events_json":"https://pith.science/api/pith-number/SH25GJLM4ZVTRFMTOJENWKADXD/events.json","paper":"https://pith.science/paper/SH25GJLM"},"agent_actions":{"view_html":"https://pith.science/pith/SH25GJLM4ZVTRFMTOJENWKADXD","download_json":"https://pith.science/pith/SH25GJLM4ZVTRFMTOJENWKADXD.json","view_paper":"https://pith.science/paper/SH25GJLM","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.11045&json=true","fetch_graph":"https://pith.science/api/pith-number/SH25GJLM4ZVTRFMTOJENWKADXD/graph.json","fetch_events":"https://pith.science/api/pith-number/SH25GJLM4ZVTRFMTOJENWKADXD/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/SH25GJLM4ZVTRFMTOJENWKADXD/action/timestamp_anchor","attest_storage":"https://pith.science/pith/SH25GJLM4ZVTRFMTOJENWKADXD/action/storage_attestation","attest_author":"https://pith.science/pith/SH25GJLM4ZVTRFMTOJENWKADXD/action/author_attestation","sign_citation":"https://pith.science/pith/SH25GJLM4ZVTRFMTOJENWKADXD/action/citation_signature","submit_replication":"https://pith.science/pith/SH25GJLM4ZVTRFMTOJENWKADXD/action/replication_record"}},"created_at":"2026-06-10T01:11:06.659047+00:00","updated_at":"2026-06-10T01:11:06.659047+00:00"}