{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:4AGCKVPO4ZZECCAETM5CL4NMLL","short_pith_number":"pith:4AGCKVPO","schema_version":"1.0","canonical_sha256":"e00c2555eee6724108049b3a25f1ac5aed4f6e2d57089c1f2183192d455c26fa","source":{"kind":"arxiv","id":"2312.01037","version":4},"attestation_state":"computed","paper":{"title":"Eliciting Latent Knowledge from Quirky Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Alex Mallen, Julia Kharchenko, Madeline Brumley, Nora Belrose","submitted_at":"2023-12-02T05:47:22Z","abstract_excerpt":"Eliciting Latent Knowledge (ELK) aims to find patterns in a capable neural network's activations that robustly track the true state of the world, especially in hard-to-verify cases where the model's output is untrusted. To further ELK research, we introduce 12 datasets and a corresponding suite of \"quirky\" language models (LMs) that are finetuned to make systematic errors when answering questions if and only if the keyword \"Bob\" is present in the prompt. We find that, especially in middle layers, linear probes usually report an LM's knowledge independently of what the LM outputs, enabling us t"},"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":"2312.01037","kind":"arxiv","version":4},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2023-12-02T05:47:22Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"ef924b98fa61e63f8754fc0fb2b38d4eda866019b9b326f96c48355f83a2342e","abstract_canon_sha256":"eb8bb47e4027f4c470844dfb80ad53422d42a8fc20114a975d86593318653adb"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:53:52.568717Z","signature_b64":"7xRDeb/6B3h8WetwU7XagtL7JTTZ0tcJoPsDS3MUCmFbkCZXHX5qJbkwjB0FHRc1pWJ6R65VTm7BCQ2qXHWvBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"e00c2555eee6724108049b3a25f1ac5aed4f6e2d57089c1f2183192d455c26fa","last_reissued_at":"2026-07-05T08:53:52.568262Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:53:52.568262Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Eliciting Latent Knowledge from Quirky Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Alex Mallen, Julia Kharchenko, Madeline Brumley, Nora Belrose","submitted_at":"2023-12-02T05:47:22Z","abstract_excerpt":"Eliciting Latent Knowledge (ELK) aims to find patterns in a capable neural network's activations that robustly track the true state of the world, especially in hard-to-verify cases where the model's output is untrusted. To further ELK research, we introduce 12 datasets and a corresponding suite of \"quirky\" language models (LMs) that are finetuned to make systematic errors when answering questions if and only if the keyword \"Bob\" is present in the prompt. We find that, especially in middle layers, linear probes usually report an LM's knowledge independently of what the LM outputs, enabling us t"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2312.01037","kind":"arxiv","version":4},"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/2312.01037/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":"2312.01037","created_at":"2026-07-05T08:53:52.568320+00:00"},{"alias_kind":"arxiv_version","alias_value":"2312.01037v4","created_at":"2026-07-05T08:53:52.568320+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2312.01037","created_at":"2026-07-05T08:53:52.568320+00:00"},{"alias_kind":"pith_short_12","alias_value":"4AGCKVPO4ZZE","created_at":"2026-07-05T08:53:52.568320+00:00"},{"alias_kind":"pith_short_16","alias_value":"4AGCKVPO4ZZECCAE","created_at":"2026-07-05T08:53:52.568320+00:00"},{"alias_kind":"pith_short_8","alias_value":"4AGCKVPO","created_at":"2026-07-05T08:53:52.568320+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":7,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.26523","citing_title":"Radical AI Interpretability","ref_index":81,"is_internal_anchor":false},{"citing_arxiv_id":"2607.01774","citing_title":"Subliminal Clocks: Latent Time Modelling in Diffusion Language Models","ref_index":23,"is_internal_anchor":false},{"citing_arxiv_id":"2606.12268","citing_title":"The Impossibility of Eliciting Latent Knowledge","ref_index":24,"is_internal_anchor":false},{"citing_arxiv_id":"2605.29358","citing_title":"Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet","ref_index":49,"is_internal_anchor":false},{"citing_arxiv_id":"2503.11926","citing_title":"Monitoring Reasoning Models for Misbehavior and the Risks of Promoting Obfuscation","ref_index":66,"is_internal_anchor":false},{"citing_arxiv_id":"2506.00166","citing_title":"Disentangled Safety Adapters Enable Efficient Guardrails and Flexible Inference-Time Alignment","ref_index":28,"is_internal_anchor":false},{"citing_arxiv_id":"2605.11448","citing_title":"Deep Minds and Shallow Probes","ref_index":29,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/4AGCKVPO4ZZECCAETM5CL4NMLL","json":"https://pith.science/pith/4AGCKVPO4ZZECCAETM5CL4NMLL.json","graph_json":"https://pith.science/api/pith-number/4AGCKVPO4ZZECCAETM5CL4NMLL/graph.json","events_json":"https://pith.science/api/pith-number/4AGCKVPO4ZZECCAETM5CL4NMLL/events.json","paper":"https://pith.science/paper/4AGCKVPO"},"agent_actions":{"view_html":"https://pith.science/pith/4AGCKVPO4ZZECCAETM5CL4NMLL","download_json":"https://pith.science/pith/4AGCKVPO4ZZECCAETM5CL4NMLL.json","view_paper":"https://pith.science/paper/4AGCKVPO","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2312.01037&json=true","fetch_graph":"https://pith.science/api/pith-number/4AGCKVPO4ZZECCAETM5CL4NMLL/graph.json","fetch_events":"https://pith.science/api/pith-number/4AGCKVPO4ZZECCAETM5CL4NMLL/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/4AGCKVPO4ZZECCAETM5CL4NMLL/action/timestamp_anchor","attest_storage":"https://pith.science/pith/4AGCKVPO4ZZECCAETM5CL4NMLL/action/storage_attestation","attest_author":"https://pith.science/pith/4AGCKVPO4ZZECCAETM5CL4NMLL/action/author_attestation","sign_citation":"https://pith.science/pith/4AGCKVPO4ZZECCAETM5CL4NMLL/action/citation_signature","submit_replication":"https://pith.science/pith/4AGCKVPO4ZZECCAETM5CL4NMLL/action/replication_record"}},"created_at":"2026-07-05T08:53:52.568320+00:00","updated_at":"2026-07-05T08:53:52.568320+00:00"}