{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:WLK5QSX5LFFCZKD2FMM7ICZY7L","short_pith_number":"pith:WLK5QSX5","schema_version":"1.0","canonical_sha256":"b2d5d84afd594a2ca87a2b19f40b38fafbe97101b1b7f1db764e7abe14f2dedd","source":{"kind":"arxiv","id":"2401.00595","version":3},"attestation_state":"computed","paper":{"title":"State of What Art? A Call for Multi-Prompt LLM Evaluation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Dafna Shahaf, Dan Malkin, Gabriel Stanovsky, Guy Kaplan, Moran Mizrahi, Rotem Dror","submitted_at":"2023-12-31T22:21:36Z","abstract_excerpt":"Recent advances in large language models (LLMs) have led to the development of various evaluation benchmarks. These benchmarks typically rely on a single instruction template for evaluating all LLMs on a specific task. In this paper, we comprehensively analyze the brittleness of results obtained via single-prompt evaluations across 6.5M instances, involving 20 different LLMs and 39 tasks from 3 benchmarks. To improve robustness of the analysis, we propose to evaluate LLMs with a set of diverse prompts instead. We discuss tailored evaluation metrics for specific use cases (e.g., LLM developers "},"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":"2401.00595","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2023-12-31T22:21:36Z","cross_cats_sorted":[],"title_canon_sha256":"bad5475ee3dd3a4489726e40eafceb84e2c456d2e82c7101a53e0b2ba4f62c4f","abstract_canon_sha256":"fdfd55e2e2a4cbb5a65bfb571d1291a59551cc5dcaedabd8b59603b56605d6a4"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:15:43.537811Z","signature_b64":"NSp4biaXvU3Qi0peL5C4+aHO5CF+I+8tfVOE5gvWqNOPyXdYgIk+i6bo3fQcX4tKjnejtbRyxKYezYRxwGnKDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"b2d5d84afd594a2ca87a2b19f40b38fafbe97101b1b7f1db764e7abe14f2dedd","last_reissued_at":"2026-07-05T08:15:43.537271Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:15:43.537271Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"State of What Art? A Call for Multi-Prompt LLM Evaluation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Dafna Shahaf, Dan Malkin, Gabriel Stanovsky, Guy Kaplan, Moran Mizrahi, Rotem Dror","submitted_at":"2023-12-31T22:21:36Z","abstract_excerpt":"Recent advances in large language models (LLMs) have led to the development of various evaluation benchmarks. These benchmarks typically rely on a single instruction template for evaluating all LLMs on a specific task. In this paper, we comprehensively analyze the brittleness of results obtained via single-prompt evaluations across 6.5M instances, involving 20 different LLMs and 39 tasks from 3 benchmarks. To improve robustness of the analysis, we propose to evaluate LLMs with a set of diverse prompts instead. We discuss tailored evaluation metrics for specific use cases (e.g., LLM developers "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2401.00595","kind":"arxiv","version":3},"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/2401.00595/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":"2401.00595","created_at":"2026-07-05T08:15:43.537343+00:00"},{"alias_kind":"arxiv_version","alias_value":"2401.00595v3","created_at":"2026-07-05T08:15:43.537343+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2401.00595","created_at":"2026-07-05T08:15:43.537343+00:00"},{"alias_kind":"pith_short_12","alias_value":"WLK5QSX5LFFC","created_at":"2026-07-05T08:15:43.537343+00:00"},{"alias_kind":"pith_short_16","alias_value":"WLK5QSX5LFFCZKD2","created_at":"2026-07-05T08:15:43.537343+00:00"},{"alias_kind":"pith_short_8","alias_value":"WLK5QSX5","created_at":"2026-07-05T08:15:43.537343+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":10,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.21937","citing_title":"Latent Confidence Alignment for LLM Self-Assessment","ref_index":31,"is_internal_anchor":false},{"citing_arxiv_id":"2606.09809","citing_title":"Evaluation Cards: An Interpretive Layer for AI Evaluation Reporting","ref_index":77,"is_internal_anchor":false},{"citing_arxiv_id":"2605.25492","citing_title":"SafetyRepro: Configuration-Conditional Rank Instability on Alignment Benchmarks","ref_index":37,"is_internal_anchor":false},{"citing_arxiv_id":"2310.11113","citing_title":"Revisiting Sentiment Analysis for Software Engineering in the Era of Large Language Models","ref_index":40,"is_internal_anchor":false},{"citing_arxiv_id":"2404.18923","citing_title":"Holmes: A Benchmark to Assess the Linguistic Competence of Language Models","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2605.21515","citing_title":"Predicting Performance of Symbolic and Prompt Programs with Examples","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2405.14782","citing_title":"Lessons from the Trenches on Reproducible Evaluation of Language Models","ref_index":137,"is_internal_anchor":false},{"citing_arxiv_id":"2401.03065","citing_title":"CRUXEval: A Benchmark for Code Reasoning, Understanding and Execution","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2403.07974","citing_title":"LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for Code","ref_index":106,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06327","citing_title":"Measuring Evaluation-Context Divergence in Open-Weight LLMs: A Paired-Prompt Protocol with Pilot Evidence of Alignment-Pipeline-Specific Heterogeneity","ref_index":14,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/WLK5QSX5LFFCZKD2FMM7ICZY7L","json":"https://pith.science/pith/WLK5QSX5LFFCZKD2FMM7ICZY7L.json","graph_json":"https://pith.science/api/pith-number/WLK5QSX5LFFCZKD2FMM7ICZY7L/graph.json","events_json":"https://pith.science/api/pith-number/WLK5QSX5LFFCZKD2FMM7ICZY7L/events.json","paper":"https://pith.science/paper/WLK5QSX5"},"agent_actions":{"view_html":"https://pith.science/pith/WLK5QSX5LFFCZKD2FMM7ICZY7L","download_json":"https://pith.science/pith/WLK5QSX5LFFCZKD2FMM7ICZY7L.json","view_paper":"https://pith.science/paper/WLK5QSX5","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2401.00595&json=true","fetch_graph":"https://pith.science/api/pith-number/WLK5QSX5LFFCZKD2FMM7ICZY7L/graph.json","fetch_events":"https://pith.science/api/pith-number/WLK5QSX5LFFCZKD2FMM7ICZY7L/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/WLK5QSX5LFFCZKD2FMM7ICZY7L/action/timestamp_anchor","attest_storage":"https://pith.science/pith/WLK5QSX5LFFCZKD2FMM7ICZY7L/action/storage_attestation","attest_author":"https://pith.science/pith/WLK5QSX5LFFCZKD2FMM7ICZY7L/action/author_attestation","sign_citation":"https://pith.science/pith/WLK5QSX5LFFCZKD2FMM7ICZY7L/action/citation_signature","submit_replication":"https://pith.science/pith/WLK5QSX5LFFCZKD2FMM7ICZY7L/action/replication_record"}},"created_at":"2026-07-05T08:15:43.537343+00:00","updated_at":"2026-07-05T08:15:43.537343+00:00"}