{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:UHXFDEJWIC2UDGHU4WG4TOBCL2","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"217074b293386fa91854a82f0b7692a482d477bf22e4e3dde7079bd30490fa69","cross_cats_sorted":["cs.AI","cs.CL","cs.LG"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.NE","submitted_at":"2026-05-10T22:00:15Z","title_canon_sha256":"e9dc7ee6fa6dcd19345458b189f00241e0bab47d5c04212ea8c44fb2f527aa93"},"schema_version":"1.0","source":{"id":"2605.09781","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.09781","created_at":"2026-06-23T01:12:08Z"},{"alias_kind":"arxiv_version","alias_value":"2605.09781v2","created_at":"2026-06-23T01:12:08Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.09781","created_at":"2026-06-23T01:12:08Z"},{"alias_kind":"pith_short_12","alias_value":"UHXFDEJWIC2U","created_at":"2026-06-23T01:12:08Z"},{"alias_kind":"pith_short_16","alias_value":"UHXFDEJWIC2UDGHU","created_at":"2026-06-23T01:12:08Z"},{"alias_kind":"pith_short_8","alias_value":"UHXFDEJW","created_at":"2026-06-23T01:12:08Z"}],"graph_snapshots":[{"event_id":"sha256:d5d8fbf156939e1e0ca10be60f26540b8eb9323634b00e1937fb3e991f157965","target":"graph","created_at":"2026-06-23T01:12:08Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"On HumanEval (164 problems), MBPP, and creative writing benchmarks, QD-LLM achieves 46.4% higher coverage and 41.4% higher QD-Score than QDAIF (p<0.001, 30 runs, Vargha-Delaney A=0.94)."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That hybrid semantic-plus-explicit behavior descriptors remain sufficiently independent (NMI = 0.08 ± 0.02) to support the formal coverage bounds of Theorem 1 and that prompt embeddings of ~32K parameters can reliably steer 70B+ frozen LLMs across the tested domains."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"QD-LLM evolves prompt embeddings via neuroevolution in a quality-diversity framework, delivering 46% higher coverage and 41% higher QD-score than prior methods on coding and writing benchmarks."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Evolving compact prompt embeddings inside frozen large language models produces more diverse outputs than standard methods."}],"snapshot_sha256":"c27d8d8b4888a348b357dd26afcfc826d8fabcaa81ad31373eb4045012101b5a"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[{"findings_count":0,"name":"claim_evidence","ran_at":"2026-05-20T07:02:01.406459Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"ai_meta_artifact","ran_at":"2026-05-19T16:36:27.365646Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"doi_title_agreement","ran_at":"2026-05-19T12:31:17.892405Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"doi_compliance","ran_at":"2026-05-19T09:58:13.393502Z","status":"completed","version":"1.0.0"}],"endpoint":"/pith/2605.09781/integrity.json","findings":[],"snapshot_sha256":"03f6d1de9ff5e7e33a870aea5a7e7e1cc87ea1bbdbee5713065bd8eaef1b9f54","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Large Language Models exhibit mode collapse, producing homogeneous outputs that fail to explore valid solution spaces. We present QD-LLM, a framework for parameter-efficient neuroevolution that evolves prompt embeddings, compact neural interfaces (~32K parameters) that steer generation in frozen LLMs (70B+ parameters), within a Quality-Diversity (QD) optimization framework. Our contributions: (1) evolved prompt embeddings via gradient-free optimization enabling behavioral steering without model fine-tuning; (2) hybrid behavior characterization combining semantic and explicit features with form","authors_text":"Dongxin Guo, Jikun Wu, Siu Ming Yiu","cross_cats":["cs.AI","cs.CL","cs.LG"],"headline":"Evolving compact prompt embeddings inside frozen large language models produces more diverse outputs than standard methods.","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.NE","submitted_at":"2026-05-10T22:00:15Z","title":"Parameter-Efficient Neuroevolution for Diverse LLM Generation: Quality-Diversity Optimization via Prompt Embedding Evolution"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.09781","kind":"arxiv","version":2},"verdict":{"created_at":"2026-05-12T03:34:01.371350Z","id":"8365227d-ebc1-43e6-882b-f1a5b1d573cc","model_set":{"reader":"grok-4.3"},"one_line_summary":"QD-LLM evolves prompt embeddings via neuroevolution in a quality-diversity framework, delivering 46% higher coverage and 41% higher QD-score than prior methods on coding and writing benchmarks.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Evolving compact prompt embeddings inside frozen large language models produces more diverse outputs than standard methods.","strongest_claim":"On HumanEval (164 problems), MBPP, and creative writing benchmarks, QD-LLM achieves 46.4% higher coverage and 41.4% higher QD-Score than QDAIF (p<0.001, 30 runs, Vargha-Delaney A=0.94).","weakest_assumption":"That hybrid semantic-plus-explicit behavior descriptors remain sufficiently independent (NMI = 0.08 ± 0.02) to support the formal coverage bounds of Theorem 1 and that prompt embeddings of ~32K parameters can reliably steer 70B+ frozen LLMs across the tested domains."}},"verdict_id":"8365227d-ebc1-43e6-882b-f1a5b1d573cc"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:6e02801bded59606a60a475fa59388454cc1468ab6639dc23df83772bed78540","target":"record","created_at":"2026-06-23T01:12:08Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"217074b293386fa91854a82f0b7692a482d477bf22e4e3dde7079bd30490fa69","cross_cats_sorted":["cs.AI","cs.CL","cs.LG"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.NE","submitted_at":"2026-05-10T22:00:15Z","title_canon_sha256":"e9dc7ee6fa6dcd19345458b189f00241e0bab47d5c04212ea8c44fb2f527aa93"},"schema_version":"1.0","source":{"id":"2605.09781","kind":"arxiv","version":2}},"canonical_sha256":"a1ee51913640b54198f4e58dc9b8225ebc9cc008f8a78a4579883a3cfeae8003","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"a1ee51913640b54198f4e58dc9b8225ebc9cc008f8a78a4579883a3cfeae8003","first_computed_at":"2026-06-23T01:12:08.168300Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-06-23T01:12:08.168300Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"FJN/6/OD6bij1c8FGXL3huBC0JY3XEkNNbE5wo2rZJZwO6qDHp7uz0c+dTL6BXXHdiZjlrFIBUVSBtpWCcA3Cg==","signature_status":"signed_v1","signed_at":"2026-06-23T01:12:08.168768Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.09781","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:6e02801bded59606a60a475fa59388454cc1468ab6639dc23df83772bed78540","sha256:d5d8fbf156939e1e0ca10be60f26540b8eb9323634b00e1937fb3e991f157965"],"state_sha256":"5b4e25979b665a9df9e5da765062e6064b23b37dc0d361a8df1e8aa8c0974c4f"}