{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:FSBWGUE7ONAFX72WO4Q43GDUBE","short_pith_number":"pith:FSBWGUE7","schema_version":"1.0","canonical_sha256":"2c8363509f73405bff567721cd9874093cc0071f6cb0189dbe0384c2d6657c7e","source":{"kind":"arxiv","id":"2506.21521","version":2},"attestation_state":"computed","paper":{"title":"Potemkin Understanding in Large Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Bec Weeks, Keyon Vafa, Marina Mancoridis, Sendhil Mullainathan","submitted_at":"2025-06-26T17:41:35Z","abstract_excerpt":"Large language models (LLMs) are regularly evaluated using benchmark datasets. But what justifies making inferences about an LLM's capabilities based on its answers to a curated set of questions? This paper first introduces a formal framework to address this question. The key is to note that the benchmarks used to test LLMs -- such as AP exams -- are also those used to test people. However, this raises an implication: these benchmarks are only valid tests if LLMs misunderstand concepts in ways that mirror human misunderstandings. Otherwise, success on benchmarks only demonstrates potemkin unde"},"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":"2506.21521","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-06-26T17:41:35Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"906bb2423398070156cf4941268533f0733b2617925559f43a739efad3f6fb5f","abstract_canon_sha256":"398d59fb230b3dde5b8609d073d28e9134a69c0f6dc57fe9ef106cb5bd1be905"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:29:14.464723Z","signature_b64":"R4o+B/S9gkowJwxxaklaCEGAFpdgw6IDnI0yb6i4jEkX/GiYh6v/3aLJvDcKz7gk1yxS6yHj77ZIX7eAOvn5Cw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"2c8363509f73405bff567721cd9874093cc0071f6cb0189dbe0384c2d6657c7e","last_reissued_at":"2026-07-05T11:29:14.464122Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:29:14.464122Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Potemkin Understanding in Large Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Bec Weeks, Keyon Vafa, Marina Mancoridis, Sendhil Mullainathan","submitted_at":"2025-06-26T17:41:35Z","abstract_excerpt":"Large language models (LLMs) are regularly evaluated using benchmark datasets. But what justifies making inferences about an LLM's capabilities based on its answers to a curated set of questions? This paper first introduces a formal framework to address this question. The key is to note that the benchmarks used to test LLMs -- such as AP exams -- are also those used to test people. However, this raises an implication: these benchmarks are only valid tests if LLMs misunderstand concepts in ways that mirror human misunderstandings. Otherwise, success on benchmarks only demonstrates potemkin unde"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.21521","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/2506.21521/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":"2506.21521","created_at":"2026-07-05T11:29:14.464183+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.21521v2","created_at":"2026-07-05T11:29:14.464183+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.21521","created_at":"2026-07-05T11:29:14.464183+00:00"},{"alias_kind":"pith_short_12","alias_value":"FSBWGUE7ONAF","created_at":"2026-07-05T11:29:14.464183+00:00"},{"alias_kind":"pith_short_16","alias_value":"FSBWGUE7ONAFX72W","created_at":"2026-07-05T11:29:14.464183+00:00"},{"alias_kind":"pith_short_8","alias_value":"FSBWGUE7","created_at":"2026-07-05T11:29:14.464183+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2507.12549","citing_title":"The Serial Scaling Hypothesis","ref_index":68,"is_internal_anchor":false},{"citing_arxiv_id":"2604.25905","citing_title":"A paradox of AI fluency","ref_index":26,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/FSBWGUE7ONAFX72WO4Q43GDUBE","json":"https://pith.science/pith/FSBWGUE7ONAFX72WO4Q43GDUBE.json","graph_json":"https://pith.science/api/pith-number/FSBWGUE7ONAFX72WO4Q43GDUBE/graph.json","events_json":"https://pith.science/api/pith-number/FSBWGUE7ONAFX72WO4Q43GDUBE/events.json","paper":"https://pith.science/paper/FSBWGUE7"},"agent_actions":{"view_html":"https://pith.science/pith/FSBWGUE7ONAFX72WO4Q43GDUBE","download_json":"https://pith.science/pith/FSBWGUE7ONAFX72WO4Q43GDUBE.json","view_paper":"https://pith.science/paper/FSBWGUE7","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.21521&json=true","fetch_graph":"https://pith.science/api/pith-number/FSBWGUE7ONAFX72WO4Q43GDUBE/graph.json","fetch_events":"https://pith.science/api/pith-number/FSBWGUE7ONAFX72WO4Q43GDUBE/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/FSBWGUE7ONAFX72WO4Q43GDUBE/action/timestamp_anchor","attest_storage":"https://pith.science/pith/FSBWGUE7ONAFX72WO4Q43GDUBE/action/storage_attestation","attest_author":"https://pith.science/pith/FSBWGUE7ONAFX72WO4Q43GDUBE/action/author_attestation","sign_citation":"https://pith.science/pith/FSBWGUE7ONAFX72WO4Q43GDUBE/action/citation_signature","submit_replication":"https://pith.science/pith/FSBWGUE7ONAFX72WO4Q43GDUBE/action/replication_record"}},"created_at":"2026-07-05T11:29:14.464183+00:00","updated_at":"2026-07-05T11:29:14.464183+00:00"}