{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:C2EF77FX2CH2HQTD34J3ABXRJR","short_pith_number":"pith:C2EF77FX","schema_version":"1.0","canonical_sha256":"16885ffcb7d08fa3c263df13b006f14c5fed9c7d837acb805d084480c896b386","source":{"kind":"arxiv","id":"2507.20208","version":2},"attestation_state":"computed","paper":{"title":"From Benchmarks to Skills: Low-Rank Factors for LLM Evaluation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Amir DN Cohen, Aviya Maimon, Gal Vishne, Reut Tsarfaty, Shauli Ravfogel","submitted_at":"2025-07-27T10:11:16Z","abstract_excerpt":"Current evaluations of large language models (LLMs) rely heavily on a growing collection of benchmarks and on aggregate benchmark scores, yet it remains unclear what this comparison actually captures, and what these scores reveal about models' underlying capabilities. Here, we propose a new paradigm for LLM evaluation, by asking whether benchmark performance reflects many independent abilities, or rather relies on a small number of shared dimensions. To answer this, we apply Factor Analysis (FA) to a massive performance matrix of LLMs versus benchmarks \\((60\\times44)\\) revealing an \\emph{intri"},"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":"2507.20208","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-07-27T10:11:16Z","cross_cats_sorted":[],"title_canon_sha256":"6caf85026ba18fcef632b3cebbfdc88805bbf3b837746a525aa8c83b943acad1","abstract_canon_sha256":"689b97c55a68fde140f7e1737b2ce21d2e4723d0783dbbaff5a1c4ac709980e7"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-12T01:09:11.233426Z","signature_b64":"vGrZDpw/hXmWfE5Y/giK/MI7e6ULFmNTrAqM39Ehp3bmOnctvR1ci5at3dftw48GC44cPQVwt94QPpIDBq6FCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"16885ffcb7d08fa3c263df13b006f14c5fed9c7d837acb805d084480c896b386","last_reissued_at":"2026-06-12T01:09:11.232448Z","signature_status":"signed_v1","first_computed_at":"2026-06-12T01:09:11.232448Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"From Benchmarks to Skills: Low-Rank Factors for LLM Evaluation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Amir DN Cohen, Aviya Maimon, Gal Vishne, Reut Tsarfaty, Shauli Ravfogel","submitted_at":"2025-07-27T10:11:16Z","abstract_excerpt":"Current evaluations of large language models (LLMs) rely heavily on a growing collection of benchmarks and on aggregate benchmark scores, yet it remains unclear what this comparison actually captures, and what these scores reveal about models' underlying capabilities. Here, we propose a new paradigm for LLM evaluation, by asking whether benchmark performance reflects many independent abilities, or rather relies on a small number of shared dimensions. To answer this, we apply Factor Analysis (FA) to a massive performance matrix of LLMs versus benchmarks \\((60\\times44)\\) revealing an \\emph{intri"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2507.20208","kind":"arxiv","version":2},"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/2507.20208/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":"2507.20208","created_at":"2026-06-12T01:09:11.232590+00:00"},{"alias_kind":"arxiv_version","alias_value":"2507.20208v2","created_at":"2026-06-12T01:09:11.232590+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2507.20208","created_at":"2026-06-12T01:09:11.232590+00:00"},{"alias_kind":"pith_short_12","alias_value":"C2EF77FX2CH2","created_at":"2026-06-12T01:09:11.232590+00:00"},{"alias_kind":"pith_short_16","alias_value":"C2EF77FX2CH2HQTD","created_at":"2026-06-12T01:09:11.232590+00:00"},{"alias_kind":"pith_short_8","alias_value":"C2EF77FX","created_at":"2026-06-12T01:09:11.232590+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":2,"sample":[{"citing_arxiv_id":"2604.12843","citing_title":"Growing Pains: Extensible and Efficient LLM Benchmarking Via Fixed Parameter Calibration","ref_index":13,"is_internal_anchor":true},{"citing_arxiv_id":"2605.06054","citing_title":"Visual Fingerprints for LLM Generation Comparison","ref_index":24,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/C2EF77FX2CH2HQTD34J3ABXRJR","json":"https://pith.science/pith/C2EF77FX2CH2HQTD34J3ABXRJR.json","graph_json":"https://pith.science/api/pith-number/C2EF77FX2CH2HQTD34J3ABXRJR/graph.json","events_json":"https://pith.science/api/pith-number/C2EF77FX2CH2HQTD34J3ABXRJR/events.json","paper":"https://pith.science/paper/C2EF77FX"},"agent_actions":{"view_html":"https://pith.science/pith/C2EF77FX2CH2HQTD34J3ABXRJR","download_json":"https://pith.science/pith/C2EF77FX2CH2HQTD34J3ABXRJR.json","view_paper":"https://pith.science/paper/C2EF77FX","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2507.20208&json=true","fetch_graph":"https://pith.science/api/pith-number/C2EF77FX2CH2HQTD34J3ABXRJR/graph.json","fetch_events":"https://pith.science/api/pith-number/C2EF77FX2CH2HQTD34J3ABXRJR/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/C2EF77FX2CH2HQTD34J3ABXRJR/action/timestamp_anchor","attest_storage":"https://pith.science/pith/C2EF77FX2CH2HQTD34J3ABXRJR/action/storage_attestation","attest_author":"https://pith.science/pith/C2EF77FX2CH2HQTD34J3ABXRJR/action/author_attestation","sign_citation":"https://pith.science/pith/C2EF77FX2CH2HQTD34J3ABXRJR/action/citation_signature","submit_replication":"https://pith.science/pith/C2EF77FX2CH2HQTD34J3ABXRJR/action/replication_record"}},"created_at":"2026-06-12T01:09:11.232590+00:00","updated_at":"2026-06-12T01:09:11.232590+00:00"}