{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:SJGN4FBPMO4AKGUCH3S7AXF7BY","short_pith_number":"pith:SJGN4FBP","schema_version":"1.0","canonical_sha256":"924cde142f63b8051a823ee5f05cbf0e311c0bfb6d4c70c5f7f5206f3fcafd61","source":{"kind":"arxiv","id":"2405.01724","version":1},"attestation_state":"computed","paper":{"title":"Large Language Models are Inconsistent and Biased Evaluators","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Dimitris Alikaniotis, Rickard Stureborg, Yoshi Suhara","submitted_at":"2024-05-02T20:42:28Z","abstract_excerpt":"The zero-shot capability of Large Language Models (LLMs) has enabled highly flexible, reference-free metrics for various tasks, making LLM evaluators common tools in NLP. However, the robustness of these LLM evaluators remains relatively understudied; existing work mainly pursued optimal performance in terms of correlating LLM scores with human expert scores. In this paper, we conduct a series of analyses using the SummEval dataset and confirm that LLMs are biased evaluators as they: (1) exhibit familiarity bias-a preference for text with lower perplexity, (2) show skewed and biased distributi"},"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":"2405.01724","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-05-02T20:42:28Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"210f4b7479ec1faade9a13d31c371340ad5fc31059309384edd2c6aea2b4c9b5","abstract_canon_sha256":"c25932d1a8dd0b93340164f0e3d6d5570f67f3571aa057b4ce05c2fb70aeae70"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:15:01.209170Z","signature_b64":"Fd0SBZeJjWATY6A1gDDocqt1xvWrHZ+JFRhSnTbrgvyyBcd4gsgwwVGx9TowA8XCVE7jDryR8jNiAnR/NDcTAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"924cde142f63b8051a823ee5f05cbf0e311c0bfb6d4c70c5f7f5206f3fcafd61","last_reissued_at":"2026-07-05T08:15:01.208754Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:15:01.208754Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Large Language Models are Inconsistent and Biased Evaluators","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Dimitris Alikaniotis, Rickard Stureborg, Yoshi Suhara","submitted_at":"2024-05-02T20:42:28Z","abstract_excerpt":"The zero-shot capability of Large Language Models (LLMs) has enabled highly flexible, reference-free metrics for various tasks, making LLM evaluators common tools in NLP. However, the robustness of these LLM evaluators remains relatively understudied; existing work mainly pursued optimal performance in terms of correlating LLM scores with human expert scores. In this paper, we conduct a series of analyses using the SummEval dataset and confirm that LLMs are biased evaluators as they: (1) exhibit familiarity bias-a preference for text with lower perplexity, (2) show skewed and biased distributi"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2405.01724","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/2405.01724/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":"2405.01724","created_at":"2026-07-05T08:15:01.208804+00:00"},{"alias_kind":"arxiv_version","alias_value":"2405.01724v1","created_at":"2026-07-05T08:15:01.208804+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2405.01724","created_at":"2026-07-05T08:15:01.208804+00:00"},{"alias_kind":"pith_short_12","alias_value":"SJGN4FBPMO4A","created_at":"2026-07-05T08:15:01.208804+00:00"},{"alias_kind":"pith_short_16","alias_value":"SJGN4FBPMO4AKGUC","created_at":"2026-07-05T08:15:01.208804+00:00"},{"alias_kind":"pith_short_8","alias_value":"SJGN4FBP","created_at":"2026-07-05T08:15:01.208804+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":22,"internal_anchor_count":2,"sample":[{"citing_arxiv_id":"2607.07436","citing_title":"The Blind Curator: How a Biased Judge Silently Disables Skill Retirement in Self-Evolving Agents","ref_index":12,"is_internal_anchor":true},{"citing_arxiv_id":"2607.08066","citing_title":"Persuasion Attacks Can Decrease Effectiveness of CoT Monitoring","ref_index":109,"is_internal_anchor":true},{"citing_arxiv_id":"2606.26185","citing_title":"Necessary but Not Sufficient: Temperature Control and Reproducibility in LLM-as-Judge Safety Evaluations","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2606.19544","citing_title":"Reliability without Validity: A Systematic, Large-Scale Evaluation of LLM-as-a-Judge Models Across Agreement, Consistency, and Bias","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2606.07951","citing_title":"From `May' to `Is': Certainty Distortion in Language Model Rewriting","ref_index":98,"is_internal_anchor":false},{"citing_arxiv_id":"2605.27988","citing_title":"Auditing Stance Asymmetry in Generative Explanations","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2605.28158","citing_title":"OR-Space: A Full-Lifecycle Workspace Benchmark for Industrial Optimization Agents","ref_index":33,"is_internal_anchor":false},{"citing_arxiv_id":"2606.20632","citing_title":"Post-Training Recipe, More Than Model Family, Shapes Multi-Agent LLM Conversational Behavior","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2502.02871","citing_title":"Position: Multimodal Large Language Models Can Significantly Advance Scientific Reasoning","ref_index":174,"is_internal_anchor":false},{"citing_arxiv_id":"2605.17998","citing_title":"Verify-Gated Completion as Admission Control in a Governed Multi-Agent Runtime: A Bounded Architecture Case Study","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2605.20312","citing_title":"Pramana: A Protocol-Layer Treatment of Claim Verification in Autonomous Agent Networks","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2605.19141","citing_title":"GRASP: Deterministic argument ranking in interaction graphs","ref_index":59,"is_internal_anchor":false},{"citing_arxiv_id":"2605.17998","citing_title":"Verify-Gated Completion as Admission Control in a Governed Multi-Agent Runtime: A Bounded Architecture Case Study","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2605.19815","citing_title":"LP-Eval: Rubric and Dataset for Measuring the Quality of Legal Proposition Generation","ref_index":25,"is_internal_anchor":false},{"citing_arxiv_id":"2410.21819","citing_title":"Self-Preference Bias in LLM-as-a-Judge","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2605.09228","citing_title":"ProactBench: Beyond What The User Asked For","ref_index":149,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08522","citing_title":"Coordinates of Capability: A Unified MTMM-Geometric Framework for LLM Evaluation","ref_index":24,"is_internal_anchor":false},{"citing_arxiv_id":"2412.05579","citing_title":"LLMs-as-Judges: A Comprehensive Survey on LLM-based Evaluation Methods","ref_index":215,"is_internal_anchor":false},{"citing_arxiv_id":"2604.24544","citing_title":"STELLAR-E: a Synthetic, Tailored, End-to-end LLM Application Rigorous Evaluator","ref_index":29,"is_internal_anchor":false},{"citing_arxiv_id":"2604.23178","citing_title":"Judging the Judges: A Systematic Evaluation of Bias Mitigation Strategies in LLM-as-a-Judge Pipelines","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2604.13717","citing_title":"On Cost-Effective LLM-as-a-Judge Improvement Techniques","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2605.02091","citing_title":"How Compliant Are GitHub Actions Workflows? A Checklist-Based Study with LLM-Assisted Auditing","ref_index":37,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/SJGN4FBPMO4AKGUCH3S7AXF7BY","json":"https://pith.science/pith/SJGN4FBPMO4AKGUCH3S7AXF7BY.json","graph_json":"https://pith.science/api/pith-number/SJGN4FBPMO4AKGUCH3S7AXF7BY/graph.json","events_json":"https://pith.science/api/pith-number/SJGN4FBPMO4AKGUCH3S7AXF7BY/events.json","paper":"https://pith.science/paper/SJGN4FBP"},"agent_actions":{"view_html":"https://pith.science/pith/SJGN4FBPMO4AKGUCH3S7AXF7BY","download_json":"https://pith.science/pith/SJGN4FBPMO4AKGUCH3S7AXF7BY.json","view_paper":"https://pith.science/paper/SJGN4FBP","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2405.01724&json=true","fetch_graph":"https://pith.science/api/pith-number/SJGN4FBPMO4AKGUCH3S7AXF7BY/graph.json","fetch_events":"https://pith.science/api/pith-number/SJGN4FBPMO4AKGUCH3S7AXF7BY/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/SJGN4FBPMO4AKGUCH3S7AXF7BY/action/timestamp_anchor","attest_storage":"https://pith.science/pith/SJGN4FBPMO4AKGUCH3S7AXF7BY/action/storage_attestation","attest_author":"https://pith.science/pith/SJGN4FBPMO4AKGUCH3S7AXF7BY/action/author_attestation","sign_citation":"https://pith.science/pith/SJGN4FBPMO4AKGUCH3S7AXF7BY/action/citation_signature","submit_replication":"https://pith.science/pith/SJGN4FBPMO4AKGUCH3S7AXF7BY/action/replication_record"}},"created_at":"2026-07-05T08:15:01.208804+00:00","updated_at":"2026-07-05T08:15:01.208804+00:00"}