{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:LISOT6PENT6WFRA5B3EDI6TQAC","short_pith_number":"pith:LISOT6PE","schema_version":"1.0","canonical_sha256":"5a24e9f9e46cfd62c41d0ec8347a7000be63b3ba79e3c81a48c98f857f8c1e3e","source":{"kind":"arxiv","id":"2507.03120","version":1},"attestation_state":"computed","paper":{"title":"How Overconfidence in Initial Choices and Underconfidence Under Criticism Modulate Change of Mind in Large Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Andrea Banino, Benedetto De Martino, Dharshan Kumaran, Joe Heyward, Larisa Markeeva, Mrinal Mathur, Petar Velickovic, Razvan Pascanu, Simon Osindero, Stephen M Fleming, Viorica Patraucean","submitted_at":"2025-07-03T18:57:43Z","abstract_excerpt":"Large language models (LLMs) exhibit strikingly conflicting behaviors: they can appear steadfastly overconfident in their initial answers whilst at the same time being prone to excessive doubt when challenged. To investigate this apparent paradox, we developed a novel experimental paradigm, exploiting the unique ability to obtain confidence estimates from LLMs without creating memory of their initial judgments -- something impossible in human participants. We show that LLMs -- Gemma 3, GPT4o and o1-preview -- exhibit a pronounced choice-supportive bias that reinforces and boosts their estimate"},"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":"2507.03120","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-07-03T18:57:43Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"5a2bf978729821560e7cbe87656e956c5703be1760b2de4813c727327b766385","abstract_canon_sha256":"0ad48c4b9cff55974ee550f181cdee89c24d4e7a7909d73f7bd51f027a8af9bc"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:31:52.874592Z","signature_b64":"RgZgjrsI583Gcyiz0mUGAOT4Q/fhNthWy7uOmmQ9YEli3D605aFxSQdTEfEC1oC3ovSOLbMF2yHVWx/VBmUIBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"5a24e9f9e46cfd62c41d0ec8347a7000be63b3ba79e3c81a48c98f857f8c1e3e","last_reissued_at":"2026-07-05T11:31:52.874134Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:31:52.874134Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"How Overconfidence in Initial Choices and Underconfidence Under Criticism Modulate Change of Mind in Large Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Andrea Banino, Benedetto De Martino, Dharshan Kumaran, Joe Heyward, Larisa Markeeva, Mrinal Mathur, Petar Velickovic, Razvan Pascanu, Simon Osindero, Stephen M Fleming, Viorica Patraucean","submitted_at":"2025-07-03T18:57:43Z","abstract_excerpt":"Large language models (LLMs) exhibit strikingly conflicting behaviors: they can appear steadfastly overconfident in their initial answers whilst at the same time being prone to excessive doubt when challenged. To investigate this apparent paradox, we developed a novel experimental paradigm, exploiting the unique ability to obtain confidence estimates from LLMs without creating memory of their initial judgments -- something impossible in human participants. We show that LLMs -- Gemma 3, GPT4o and o1-preview -- exhibit a pronounced choice-supportive bias that reinforces and boosts their estimate"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2507.03120","kind":"arxiv","version":1},"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/2507.03120/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":"2507.03120","created_at":"2026-07-05T11:31:52.874188+00:00"},{"alias_kind":"arxiv_version","alias_value":"2507.03120v1","created_at":"2026-07-05T11:31:52.874188+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2507.03120","created_at":"2026-07-05T11:31:52.874188+00:00"},{"alias_kind":"pith_short_12","alias_value":"LISOT6PENT6W","created_at":"2026-07-05T11:31:52.874188+00:00"},{"alias_kind":"pith_short_16","alias_value":"LISOT6PENT6WFRA5","created_at":"2026-07-05T11:31:52.874188+00:00"},{"alias_kind":"pith_short_8","alias_value":"LISOT6PE","created_at":"2026-07-05T11:31:52.874188+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.31561","citing_title":"What Am I Missing? Question-Answering as Hidden State Probing","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2601.22297","citing_title":"Learning from Self-Debate: Preparing Reasoning Models for Multi-Agent Debate","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2603.22161","citing_title":"Causal Evidence that Language Models use Confidence to Drive Behavior","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2604.17707","citing_title":"Before You Interpret the Profile: Validity Scaling for LLM Metacognitive Self-Report","ref_index":8,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/LISOT6PENT6WFRA5B3EDI6TQAC","json":"https://pith.science/pith/LISOT6PENT6WFRA5B3EDI6TQAC.json","graph_json":"https://pith.science/api/pith-number/LISOT6PENT6WFRA5B3EDI6TQAC/graph.json","events_json":"https://pith.science/api/pith-number/LISOT6PENT6WFRA5B3EDI6TQAC/events.json","paper":"https://pith.science/paper/LISOT6PE"},"agent_actions":{"view_html":"https://pith.science/pith/LISOT6PENT6WFRA5B3EDI6TQAC","download_json":"https://pith.science/pith/LISOT6PENT6WFRA5B3EDI6TQAC.json","view_paper":"https://pith.science/paper/LISOT6PE","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2507.03120&json=true","fetch_graph":"https://pith.science/api/pith-number/LISOT6PENT6WFRA5B3EDI6TQAC/graph.json","fetch_events":"https://pith.science/api/pith-number/LISOT6PENT6WFRA5B3EDI6TQAC/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/LISOT6PENT6WFRA5B3EDI6TQAC/action/timestamp_anchor","attest_storage":"https://pith.science/pith/LISOT6PENT6WFRA5B3EDI6TQAC/action/storage_attestation","attest_author":"https://pith.science/pith/LISOT6PENT6WFRA5B3EDI6TQAC/action/author_attestation","sign_citation":"https://pith.science/pith/LISOT6PENT6WFRA5B3EDI6TQAC/action/citation_signature","submit_replication":"https://pith.science/pith/LISOT6PENT6WFRA5B3EDI6TQAC/action/replication_record"}},"created_at":"2026-07-05T11:31:52.874188+00:00","updated_at":"2026-07-05T11:31:52.874188+00:00"}