{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:Y7ABMYYQNBKFIWX6J26XX445BK","short_pith_number":"pith:Y7ABMYYQ","schema_version":"1.0","canonical_sha256":"c7c01663106854545afe4ebd7bf39d0a98ea44d412d56d350589393e8c00319a","source":{"kind":"arxiv","id":"2405.03205","version":3},"attestation_state":"computed","paper":{"title":"Anchored Answers: Unravelling Positional Bias in GPT-2's Multiple-Choice Questions","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Ruizhe Li, Yanjun Gao","submitted_at":"2024-05-06T07:10:09Z","abstract_excerpt":"Large Language Models (LLMs), such as the GPT-4 and LLaMA families, have demonstrated considerable success across diverse tasks, including multiple-choice questions (MCQs). However, these models exhibit a positional bias, particularly an even worse anchored bias in the GPT-2 family, where they consistently favour the first choice 'A' in MCQs during inference. This anchored bias challenges the integrity of GPT-2's decision-making process, as it skews performance based on the position rather than the content of the choices in MCQs. In this study, we utilise the mechanistic interpretability appro"},"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":"2405.03205","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-05-06T07:10:09Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"85b70684ddb81f6d2d5823d9168d7c89a799d6039c21c2ce6e63ce93c613b6d3","abstract_canon_sha256":"84cdf36d7f4eebd905bf18093649ea71d4ebfffc171596eaa3c705a7692d8d12"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:13:09.529254Z","signature_b64":"DmeZI737DT6yrJ7Ol4O2y3qb+Bc0ilcEPn4vMbmxGGTNSSOMhHa13pjho/tIen6H/gXMNCtxlhCeeiEuLWaWBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c7c01663106854545afe4ebd7bf39d0a98ea44d412d56d350589393e8c00319a","last_reissued_at":"2026-07-05T11:13:09.528696Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:13:09.528696Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Anchored Answers: Unravelling Positional Bias in GPT-2's Multiple-Choice Questions","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Ruizhe Li, Yanjun Gao","submitted_at":"2024-05-06T07:10:09Z","abstract_excerpt":"Large Language Models (LLMs), such as the GPT-4 and LLaMA families, have demonstrated considerable success across diverse tasks, including multiple-choice questions (MCQs). However, these models exhibit a positional bias, particularly an even worse anchored bias in the GPT-2 family, where they consistently favour the first choice 'A' in MCQs during inference. This anchored bias challenges the integrity of GPT-2's decision-making process, as it skews performance based on the position rather than the content of the choices in MCQs. In this study, we utilise the mechanistic interpretability appro"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2405.03205","kind":"arxiv","version":3},"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/2405.03205/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":"2405.03205","created_at":"2026-07-05T11:13:09.528764+00:00"},{"alias_kind":"arxiv_version","alias_value":"2405.03205v3","created_at":"2026-07-05T11:13:09.528764+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2405.03205","created_at":"2026-07-05T11:13:09.528764+00:00"},{"alias_kind":"pith_short_12","alias_value":"Y7ABMYYQNBKF","created_at":"2026-07-05T11:13:09.528764+00:00"},{"alias_kind":"pith_short_16","alias_value":"Y7ABMYYQNBKFIWX6","created_at":"2026-07-05T11:13:09.528764+00:00"},{"alias_kind":"pith_short_8","alias_value":"Y7ABMYYQ","created_at":"2026-07-05T11:13:09.528764+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2507.18182","citing_title":"SCOPE: Stochastic and Counterbiased Option Placement for Evaluating Large Language Models","ref_index":10,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/Y7ABMYYQNBKFIWX6J26XX445BK","json":"https://pith.science/pith/Y7ABMYYQNBKFIWX6J26XX445BK.json","graph_json":"https://pith.science/api/pith-number/Y7ABMYYQNBKFIWX6J26XX445BK/graph.json","events_json":"https://pith.science/api/pith-number/Y7ABMYYQNBKFIWX6J26XX445BK/events.json","paper":"https://pith.science/paper/Y7ABMYYQ"},"agent_actions":{"view_html":"https://pith.science/pith/Y7ABMYYQNBKFIWX6J26XX445BK","download_json":"https://pith.science/pith/Y7ABMYYQNBKFIWX6J26XX445BK.json","view_paper":"https://pith.science/paper/Y7ABMYYQ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2405.03205&json=true","fetch_graph":"https://pith.science/api/pith-number/Y7ABMYYQNBKFIWX6J26XX445BK/graph.json","fetch_events":"https://pith.science/api/pith-number/Y7ABMYYQNBKFIWX6J26XX445BK/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/Y7ABMYYQNBKFIWX6J26XX445BK/action/timestamp_anchor","attest_storage":"https://pith.science/pith/Y7ABMYYQNBKFIWX6J26XX445BK/action/storage_attestation","attest_author":"https://pith.science/pith/Y7ABMYYQNBKFIWX6J26XX445BK/action/author_attestation","sign_citation":"https://pith.science/pith/Y7ABMYYQNBKFIWX6J26XX445BK/action/citation_signature","submit_replication":"https://pith.science/pith/Y7ABMYYQNBKFIWX6J26XX445BK/action/replication_record"}},"created_at":"2026-07-05T11:13:09.528764+00:00","updated_at":"2026-07-05T11:13:09.528764+00:00"}