{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:RVVQGMDFC4VBRIWB5T3KDYCMXT","short_pith_number":"pith:RVVQGMDF","schema_version":"1.0","canonical_sha256":"8d6b033065172a18a2c1ecf6a1e04cbcd30be6024630cd0f861b321c55f21f1d","source":{"kind":"arxiv","id":"2605.30148","version":1},"attestation_state":"computed","paper":{"title":"Overcoming Forgetting in LLM Fine-Tuning with Evolution Strategies","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Conor F. Hayes, Kajetan Schweighofer, Risto Miikkulainen, Roberto Dailey, Xin Qiu","submitted_at":"2026-05-28T16:08:47Z","abstract_excerpt":"Evolution Strategies (ES) has recently emerged as a competitive alternative to reinforcement learning (RL) for large language model (LLM) fine-tuning, offering advantages through simplicity, scalability, and inference-only training. However, recent work suggests that ES fine-tuning on new tasks may induce forgetting of prior tasks. First, this paper shows that prior task forgetting (1) is better characterized as performance drift rather than irreversible forgetting, with prior-task performance often recovering during ES training; and (2) is not a specific failure mode of ES, but can also arise"},"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":"2605.30148","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-28T16:08:47Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"a9692a470d1937210d72b71b12c3cd76d69b1ad99b3dcb8a4a1cd40abc383e41","abstract_canon_sha256":"f7c8bfb7b3ccc7f53a362775603e4e6d70dd7610a7de833ff9f827a49a5521d4"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-29T02:06:11.107208Z","signature_b64":"kqkBeyzFY+bXzMswIe+QlcAD6E660Ch28oqyplt1XkbszClGa3NJ04GA3d4IKzZg3J3raEZ1rpVt0ThErAW4AA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"8d6b033065172a18a2c1ecf6a1e04cbcd30be6024630cd0f861b321c55f21f1d","last_reissued_at":"2026-05-29T02:06:11.106857Z","signature_status":"signed_v1","first_computed_at":"2026-05-29T02:06:11.106857Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Overcoming Forgetting in LLM Fine-Tuning with Evolution Strategies","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Conor F. Hayes, Kajetan Schweighofer, Risto Miikkulainen, Roberto Dailey, Xin Qiu","submitted_at":"2026-05-28T16:08:47Z","abstract_excerpt":"Evolution Strategies (ES) has recently emerged as a competitive alternative to reinforcement learning (RL) for large language model (LLM) fine-tuning, offering advantages through simplicity, scalability, and inference-only training. However, recent work suggests that ES fine-tuning on new tasks may induce forgetting of prior tasks. First, this paper shows that prior task forgetting (1) is better characterized as performance drift rather than irreversible forgetting, with prior-task performance often recovering during ES training; and (2) is not a specific failure mode of ES, but can also arise"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.30148","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/2605.30148/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":"2605.30148","created_at":"2026-05-29T02:06:11.106917+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.30148v1","created_at":"2026-05-29T02:06:11.106917+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.30148","created_at":"2026-05-29T02:06:11.106917+00:00"},{"alias_kind":"pith_short_12","alias_value":"RVVQGMDFC4VB","created_at":"2026-05-29T02:06:11.106917+00:00"},{"alias_kind":"pith_short_16","alias_value":"RVVQGMDFC4VBRIWB","created_at":"2026-05-29T02:06:11.106917+00:00"},{"alias_kind":"pith_short_8","alias_value":"RVVQGMDF","created_at":"2026-05-29T02:06:11.106917+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2606.29119","citing_title":"Knowing in Advance When an Evolutionary Outer Loop Will Not Help: A Pre-Registered Cheap-Baseline Screening Rule","ref_index":12,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/RVVQGMDFC4VBRIWB5T3KDYCMXT","json":"https://pith.science/pith/RVVQGMDFC4VBRIWB5T3KDYCMXT.json","graph_json":"https://pith.science/api/pith-number/RVVQGMDFC4VBRIWB5T3KDYCMXT/graph.json","events_json":"https://pith.science/api/pith-number/RVVQGMDFC4VBRIWB5T3KDYCMXT/events.json","paper":"https://pith.science/paper/RVVQGMDF"},"agent_actions":{"view_html":"https://pith.science/pith/RVVQGMDFC4VBRIWB5T3KDYCMXT","download_json":"https://pith.science/pith/RVVQGMDFC4VBRIWB5T3KDYCMXT.json","view_paper":"https://pith.science/paper/RVVQGMDF","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.30148&json=true","fetch_graph":"https://pith.science/api/pith-number/RVVQGMDFC4VBRIWB5T3KDYCMXT/graph.json","fetch_events":"https://pith.science/api/pith-number/RVVQGMDFC4VBRIWB5T3KDYCMXT/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/RVVQGMDFC4VBRIWB5T3KDYCMXT/action/timestamp_anchor","attest_storage":"https://pith.science/pith/RVVQGMDFC4VBRIWB5T3KDYCMXT/action/storage_attestation","attest_author":"https://pith.science/pith/RVVQGMDFC4VBRIWB5T3KDYCMXT/action/author_attestation","sign_citation":"https://pith.science/pith/RVVQGMDFC4VBRIWB5T3KDYCMXT/action/citation_signature","submit_replication":"https://pith.science/pith/RVVQGMDFC4VBRIWB5T3KDYCMXT/action/replication_record"}},"created_at":"2026-05-29T02:06:11.106917+00:00","updated_at":"2026-05-29T02:06:11.106917+00:00"}