{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:TKKL3KV2U3M6IRVLQXFQUGQPIR","short_pith_number":"pith:TKKL3KV2","schema_version":"1.0","canonical_sha256":"9a94bdaabaa6d9e446ab85cb0a1a0f44629ed683c25a0ff05c81dfd985d049e1","source":{"kind":"arxiv","id":"2510.16727","version":2},"attestation_state":"computed","paper":{"title":"Beacon: Single-Turn Diagnosis and Mitigation of Latent Sycophancy in Large Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Angkul Puniya, Ruhaan Chopra, Sanskar Pandey, Sohom Pal","submitted_at":"2025-10-19T06:36:57Z","abstract_excerpt":"Large language models internalize a structural trade-off between truthfulness and obsequious flattery, emerging from reward optimization that conflates helpfulness with polite submission. This latent bias, known as sycophancy, manifests as a preference for user agreement over principled reasoning. We introduce Beacon, a single-turn forced-choice benchmark that isolates this bias independent of conversational context, enabling precise measurement of the tension between factual accuracy and submissive bias. Evaluations across twelve state-of-the-art models reveal that sycophancy decomposes into "},"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":"2510.16727","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-10-19T06:36:57Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"698838209424d7427a9b013186884f82010997d00dc086f57f80b4f0ec2260b6","abstract_canon_sha256":"23dfe841527b3d0086a34937ba2ce7814c628cbf8545c8dcc8008c00f049971b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T00:04:15.087013Z","signature_b64":"HmfssBEvmoXs5/4cNTPu0rkcjTtegTWxrRU9qtCHsw2WrrPWMbstZWfWVwlbFu1mAkZvqPQsjfLBgipjAtL0Dg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"9a94bdaabaa6d9e446ab85cb0a1a0f44629ed683c25a0ff05c81dfd985d049e1","last_reissued_at":"2026-05-20T00:04:15.086239Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T00:04:15.086239Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Beacon: Single-Turn Diagnosis and Mitigation of Latent Sycophancy in Large Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Angkul Puniya, Ruhaan Chopra, Sanskar Pandey, Sohom Pal","submitted_at":"2025-10-19T06:36:57Z","abstract_excerpt":"Large language models internalize a structural trade-off between truthfulness and obsequious flattery, emerging from reward optimization that conflates helpfulness with polite submission. This latent bias, known as sycophancy, manifests as a preference for user agreement over principled reasoning. We introduce Beacon, a single-turn forced-choice benchmark that isolates this bias independent of conversational context, enabling precise measurement of the tension between factual accuracy and submissive bias. Evaluations across twelve state-of-the-art models reveal that sycophancy decomposes into "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2510.16727","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/2510.16727/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":"2510.16727","created_at":"2026-05-20T00:04:15.086363+00:00"},{"alias_kind":"arxiv_version","alias_value":"2510.16727v2","created_at":"2026-05-20T00:04:15.086363+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2510.16727","created_at":"2026-05-20T00:04:15.086363+00:00"},{"alias_kind":"pith_short_12","alias_value":"TKKL3KV2U3M6","created_at":"2026-05-20T00:04:15.086363+00:00"},{"alias_kind":"pith_short_16","alias_value":"TKKL3KV2U3M6IRVL","created_at":"2026-05-20T00:04:15.086363+00:00"},{"alias_kind":"pith_short_8","alias_value":"TKKL3KV2","created_at":"2026-05-20T00:04:15.086363+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":2,"sample":[{"citing_arxiv_id":"2605.18799","citing_title":"ReCrit: Transition-Aware Reinforcement Learning for Scientific Critic Reasoning","ref_index":27,"is_internal_anchor":true},{"citing_arxiv_id":"2604.13602","citing_title":"Reward Hacking in the Era of Large Models: Mechanisms, Emergent Misalignment, Challenges","ref_index":47,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/TKKL3KV2U3M6IRVLQXFQUGQPIR","json":"https://pith.science/pith/TKKL3KV2U3M6IRVLQXFQUGQPIR.json","graph_json":"https://pith.science/api/pith-number/TKKL3KV2U3M6IRVLQXFQUGQPIR/graph.json","events_json":"https://pith.science/api/pith-number/TKKL3KV2U3M6IRVLQXFQUGQPIR/events.json","paper":"https://pith.science/paper/TKKL3KV2"},"agent_actions":{"view_html":"https://pith.science/pith/TKKL3KV2U3M6IRVLQXFQUGQPIR","download_json":"https://pith.science/pith/TKKL3KV2U3M6IRVLQXFQUGQPIR.json","view_paper":"https://pith.science/paper/TKKL3KV2","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2510.16727&json=true","fetch_graph":"https://pith.science/api/pith-number/TKKL3KV2U3M6IRVLQXFQUGQPIR/graph.json","fetch_events":"https://pith.science/api/pith-number/TKKL3KV2U3M6IRVLQXFQUGQPIR/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/TKKL3KV2U3M6IRVLQXFQUGQPIR/action/timestamp_anchor","attest_storage":"https://pith.science/pith/TKKL3KV2U3M6IRVLQXFQUGQPIR/action/storage_attestation","attest_author":"https://pith.science/pith/TKKL3KV2U3M6IRVLQXFQUGQPIR/action/author_attestation","sign_citation":"https://pith.science/pith/TKKL3KV2U3M6IRVLQXFQUGQPIR/action/citation_signature","submit_replication":"https://pith.science/pith/TKKL3KV2U3M6IRVLQXFQUGQPIR/action/replication_record"}},"created_at":"2026-05-20T00:04:15.086363+00:00","updated_at":"2026-05-20T00:04:15.086363+00:00"}