{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2024:CVEHX7XFUKF5DJRZPQYQDRXULA","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":"3939e6fde3f28e3e5448035dc81ace7f4a29cf0370051bdb1e096f49993cfa03","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-10-29T07:42:18Z","title_canon_sha256":"3b78afe76893d6460aba6e3ad0163ef7cd549dbc3cff746644ae9a9fa89484a6"},"schema_version":"1.0","source":{"id":"2410.21819","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2410.21819","created_at":"2026-05-17T23:38:52Z"},{"alias_kind":"arxiv_version","alias_value":"2410.21819v2","created_at":"2026-05-17T23:38:52Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2410.21819","created_at":"2026-05-17T23:38:52Z"},{"alias_kind":"pith_short_12","alias_value":"CVEHX7XFUKF5","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"CVEHX7XFUKF5DJRZ","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"CVEHX7XF","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:54a80fcefbfefc8787391f17b31cb1dd26f92ede55d065806e741b047243a85a","target":"graph","created_at":"2026-05-17T23:38:52Z","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":"LLMs assign significantly higher evaluations to outputs with lower perplexity than human evaluators, regardless of whether the outputs were self-generated. This suggests that the essence of the bias lies in perplexity."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the introduced quantitative metric isolates self-preference bias from other confounding factors in LLM judgments and that the observed correlation with perplexity is causal rather than correlational."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"LLMs judge their own outputs higher because they assign better scores to lower-perplexity text, even when the text is not self-generated."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"LLMs as judges give higher scores to low-perplexity outputs than humans, even for non-self-generated text."}],"snapshot_sha256":"b88afe9a505f5bed9eb2fa59f15e8d2e711563d960609587b928592d4625f10f"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"cf90e20267b64a74744ee5afa5fe7365ee7ac929677e5ffcf204c56b1e214bf6"},"paper":{"abstract_excerpt":"Automated evaluation leveraging large language models (LLMs), commonly referred to as LLM evaluators or LLM-as-a-judge, has been widely used in measuring the performance of dialogue systems. However, the self-preference bias in LLMs has posed significant risks, including promoting specific styles or policies intrinsic to the LLMs. Despite the importance of this issue, there is a lack of established methods to measure the self-preference bias quantitatively, and its underlying causes are poorly understood. In this paper, we introduce a novel quantitative metric to measure the self-preference bi","authors_text":"Koki Wataoka, Ryokan Ri, Tsubasa Takahashi","cross_cats":[],"headline":"LLMs as judges give higher scores to low-perplexity outputs than humans, even for non-self-generated text.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-10-29T07:42:18Z","title":"Self-Preference Bias in LLM-as-a-Judge"},"references":{"count":19,"internal_anchors":3,"resolved_work":19,"sample":[{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":1,"title":"Daniel Deutsch, Rotem Dror, and Dan Roth","work_id":"9c5e3e50-9dfd-46b0-8bcd-68a014329574","year":2022},{"cited_arxiv_id":"","doi":"10.18653/v1/2022.emnlp-main.753","is_internal_anchor":false,"ref_index":2,"title":"On the limitations of reference-free evaluations of generated text","work_id":"78772036-6fc7-459c-9c27-78fc06e14ba3","year":2022},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"Bowman, and Shi Feng","work_id":"ec9ad2bb-47a2-46c7-b8e9-05f69142decd","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"Chenghao Yang, Sida Li, and Ari Holtzman","work_id":"ae09097d-58a5-40bc-8f92-14d333fc4a2d","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"Large language models are inconsistent and biased evaluators","work_id":"36f79a6c-dedd-407f-a53d-20491052e910","year":null}],"snapshot_sha256":"bb4713df88645d3ad40abc47c71b26be069a96e88e0a57f199ede6bfcafe082a"},"source":{"id":"2410.21819","kind":"arxiv","version":2},"verdict":{"created_at":"2026-05-15T14:41:39.613451Z","id":"0c1f9f11-198c-40b0-b960-8a279a2b9b36","model_set":{"reader":"grok-4.3"},"one_line_summary":"LLMs judge their own outputs higher because they assign better scores to lower-perplexity text, even when the text is not self-generated.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"LLMs as judges give higher scores to low-perplexity outputs than humans, even for non-self-generated text.","strongest_claim":"LLMs assign significantly higher evaluations to outputs with lower perplexity than human evaluators, regardless of whether the outputs were self-generated. This suggests that the essence of the bias lies in perplexity.","weakest_assumption":"That the introduced quantitative metric isolates self-preference bias from other confounding factors in LLM judgments and that the observed correlation with perplexity is causal rather than correlational."}},"verdict_id":"0c1f9f11-198c-40b0-b960-8a279a2b9b36"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:656d742bed35d394ac3fb8d439b04aea8b2256530b76a383e81fbceece64f00f","target":"record","created_at":"2026-05-17T23:38:52Z","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":"3939e6fde3f28e3e5448035dc81ace7f4a29cf0370051bdb1e096f49993cfa03","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-10-29T07:42:18Z","title_canon_sha256":"3b78afe76893d6460aba6e3ad0163ef7cd549dbc3cff746644ae9a9fa89484a6"},"schema_version":"1.0","source":{"id":"2410.21819","kind":"arxiv","version":2}},"canonical_sha256":"15487bfee5a28bd1a6397c3101c6f45813b6d8a765e6b9054f054e2bec2ead97","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"15487bfee5a28bd1a6397c3101c6f45813b6d8a765e6b9054f054e2bec2ead97","first_computed_at":"2026-05-17T23:38:52.319961Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-17T23:38:52.319961Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"gFqF/U/2ogdpWmM9X2TD073OTauYLhRidoge9M5t0zo3jkR8Wgn9iTZcoP4Dip0pOEDvB7W8+dw/mfNX+Y9RDA==","signature_status":"signed_v1","signed_at":"2026-05-17T23:38:52.320468Z","signed_message":"canonical_sha256_bytes"},"source_id":"2410.21819","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:656d742bed35d394ac3fb8d439b04aea8b2256530b76a383e81fbceece64f00f","sha256:54a80fcefbfefc8787391f17b31cb1dd26f92ede55d065806e741b047243a85a"],"state_sha256":"258c0ccbbe1ef97bb11e4494e1390ac1ef6d9760eac75ab2d55e30a1f44969e5"}