{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:QNHVG6UPZLXU2LG5LPU7RXVDGV","short_pith_number":"pith:QNHVG6UP","schema_version":"1.0","canonical_sha256":"834f537a8fcaef4d2cdd5be9f8dea3357d8aa81d4c263becfa0b9192ddb2afd2","source":{"kind":"arxiv","id":"2512.20638","version":2},"attestation_state":"computed","paper":{"title":"Uncovering Competency Gaps in Large Language Models and Their Benchmarks","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Christoph Bregler, Maty Bohacek, Nicholas Dufour, Nino Scherrer, Stephanie C. Y. Chan, Thomas Leung","submitted_at":"2025-12-06T17:39:47Z","abstract_excerpt":"The evaluation of large language models relies heavily on standardized benchmarks. These benchmarks provide useful aggregated metrics, but can obscure (i) particular sub-areas where the models are weak (\"model gaps\") and (ii) imbalanced coverage in the benchmarks themselves (\"benchmark gaps\"). To automatically uncover both types of gaps, we propose a simple new method using concept activations from sparse autoencoders, to identify fine-grained gaps on a per-concept basis. The method also benefits from grounding evaluation in the model's internal representations, as well as easy comparison acro"},"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":"2512.20638","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-12-06T17:39:47Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"63baa7e85a487c1a3f0b336804c016458574fa6fcea48b2359d5022c80e9fc8c","abstract_canon_sha256":"b1f797b793b4733c8a4502b9ae89ad686cedabd9197a249138338d89ac9545e7"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-02T01:03:40.163977Z","signature_b64":"WUehgjt78HQ0ZVEVJM/VxQFFEri3I8s0NKvrb54wxGT6KwvRVBL3AToutmkEKKEx3pt2qJJbewTFOL/H4MWGDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"834f537a8fcaef4d2cdd5be9f8dea3357d8aa81d4c263becfa0b9192ddb2afd2","last_reissued_at":"2026-06-02T01:03:40.163441Z","signature_status":"signed_v1","first_computed_at":"2026-06-02T01:03:40.163441Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Uncovering Competency Gaps in Large Language Models and Their Benchmarks","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Christoph Bregler, Maty Bohacek, Nicholas Dufour, Nino Scherrer, Stephanie C. Y. Chan, Thomas Leung","submitted_at":"2025-12-06T17:39:47Z","abstract_excerpt":"The evaluation of large language models relies heavily on standardized benchmarks. These benchmarks provide useful aggregated metrics, but can obscure (i) particular sub-areas where the models are weak (\"model gaps\") and (ii) imbalanced coverage in the benchmarks themselves (\"benchmark gaps\"). To automatically uncover both types of gaps, we propose a simple new method using concept activations from sparse autoencoders, to identify fine-grained gaps on a per-concept basis. The method also benefits from grounding evaluation in the model's internal representations, as well as easy comparison acro"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2512.20638","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/2512.20638/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":"2512.20638","created_at":"2026-06-02T01:03:40.163503+00:00"},{"alias_kind":"arxiv_version","alias_value":"2512.20638v2","created_at":"2026-06-02T01:03:40.163503+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2512.20638","created_at":"2026-06-02T01:03:40.163503+00:00"},{"alias_kind":"pith_short_12","alias_value":"QNHVG6UPZLXU","created_at":"2026-06-02T01:03:40.163503+00:00"},{"alias_kind":"pith_short_16","alias_value":"QNHVG6UPZLXU2LG5","created_at":"2026-06-02T01:03:40.163503+00:00"},{"alias_kind":"pith_short_8","alias_value":"QNHVG6UP","created_at":"2026-06-02T01:03:40.163503+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2605.14164","citing_title":"Unsteady Metrics and Benchmarking Cultures of AI Model Builders","ref_index":7,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/QNHVG6UPZLXU2LG5LPU7RXVDGV","json":"https://pith.science/pith/QNHVG6UPZLXU2LG5LPU7RXVDGV.json","graph_json":"https://pith.science/api/pith-number/QNHVG6UPZLXU2LG5LPU7RXVDGV/graph.json","events_json":"https://pith.science/api/pith-number/QNHVG6UPZLXU2LG5LPU7RXVDGV/events.json","paper":"https://pith.science/paper/QNHVG6UP"},"agent_actions":{"view_html":"https://pith.science/pith/QNHVG6UPZLXU2LG5LPU7RXVDGV","download_json":"https://pith.science/pith/QNHVG6UPZLXU2LG5LPU7RXVDGV.json","view_paper":"https://pith.science/paper/QNHVG6UP","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2512.20638&json=true","fetch_graph":"https://pith.science/api/pith-number/QNHVG6UPZLXU2LG5LPU7RXVDGV/graph.json","fetch_events":"https://pith.science/api/pith-number/QNHVG6UPZLXU2LG5LPU7RXVDGV/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/QNHVG6UPZLXU2LG5LPU7RXVDGV/action/timestamp_anchor","attest_storage":"https://pith.science/pith/QNHVG6UPZLXU2LG5LPU7RXVDGV/action/storage_attestation","attest_author":"https://pith.science/pith/QNHVG6UPZLXU2LG5LPU7RXVDGV/action/author_attestation","sign_citation":"https://pith.science/pith/QNHVG6UPZLXU2LG5LPU7RXVDGV/action/citation_signature","submit_replication":"https://pith.science/pith/QNHVG6UPZLXU2LG5LPU7RXVDGV/action/replication_record"}},"created_at":"2026-06-02T01:03:40.163503+00:00","updated_at":"2026-06-02T01:03:40.163503+00:00"}