{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:EJWQIF3XKWIBIDTBC66CCBQ7LD","short_pith_number":"pith:EJWQIF3X","schema_version":"1.0","canonical_sha256":"226d0417775590140e6117bc21061f58d02194edafd5321ec31b53439a789c71","source":{"kind":"arxiv","id":"2411.05345","version":1},"attestation_state":"computed","paper":{"title":"Reasoning Robustness of LLMs to Adversarial Typographical Errors","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Anirudh Goyal, Esther Gan, Kenji Kawaguchi, Liying Cheng, Michael Shieh, Min-Yen Kan, Yancan Mao, Yiran Zhao","submitted_at":"2024-11-08T05:54:05Z","abstract_excerpt":"Large Language Models (LLMs) have demonstrated impressive capabilities in reasoning using Chain-of-Thought (CoT) prompting. However, CoT can be biased by users' instruction. In this work, we study the reasoning robustness of LLMs to typographical errors, which can naturally occur in users' queries. We design an Adversarial Typo Attack ($\\texttt{ATA}$) algorithm that iteratively samples typos for words that are important to the query and selects the edit that is most likely to succeed in attacking. It shows that LLMs are sensitive to minimal adversarial typographical changes. Notably, with 1 ch"},"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":"2411.05345","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-11-08T05:54:05Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"fee4465c60458370359a1ea53dcbdb2ebde6b2533eeea3941da8fae58f6becbe","abstract_canon_sha256":"6e4c939e15933f783dafaad2f4cfba1294446cc0f834f97f8b6727e1979850bf"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:32:59.007777Z","signature_b64":"Lq8g7lZRKeoE9X3fVHPSWf7swESjACfmDwwdg3KfTSI3jhBV1A/28UTKGOP3A67DGZ7rv+wNOJX2DiVtOm2ACQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"226d0417775590140e6117bc21061f58d02194edafd5321ec31b53439a789c71","last_reissued_at":"2026-07-05T09:32:59.007256Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:32:59.007256Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Reasoning Robustness of LLMs to Adversarial Typographical Errors","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Anirudh Goyal, Esther Gan, Kenji Kawaguchi, Liying Cheng, Michael Shieh, Min-Yen Kan, Yancan Mao, Yiran Zhao","submitted_at":"2024-11-08T05:54:05Z","abstract_excerpt":"Large Language Models (LLMs) have demonstrated impressive capabilities in reasoning using Chain-of-Thought (CoT) prompting. However, CoT can be biased by users' instruction. In this work, we study the reasoning robustness of LLMs to typographical errors, which can naturally occur in users' queries. We design an Adversarial Typo Attack ($\\texttt{ATA}$) algorithm that iteratively samples typos for words that are important to the query and selects the edit that is most likely to succeed in attacking. It shows that LLMs are sensitive to minimal adversarial typographical changes. Notably, with 1 ch"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2411.05345","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/2411.05345/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":"2411.05345","created_at":"2026-07-05T09:32:59.007320+00:00"},{"alias_kind":"arxiv_version","alias_value":"2411.05345v1","created_at":"2026-07-05T09:32:59.007320+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2411.05345","created_at":"2026-07-05T09:32:59.007320+00:00"},{"alias_kind":"pith_short_12","alias_value":"EJWQIF3XKWIB","created_at":"2026-07-05T09:32:59.007320+00:00"},{"alias_kind":"pith_short_16","alias_value":"EJWQIF3XKWIBIDTB","created_at":"2026-07-05T09:32:59.007320+00:00"},{"alias_kind":"pith_short_8","alias_value":"EJWQIF3X","created_at":"2026-07-05T09:32:59.007320+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.26396","citing_title":"At the Edge of Understanding: Sparse Autoencoders Trace The Limits of Transformer Generalization","ref_index":33,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10516","citing_title":"Consistency as a Testable Property: Statistical Methods to Evaluate AI Agent Reliability","ref_index":5,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/EJWQIF3XKWIBIDTBC66CCBQ7LD","json":"https://pith.science/pith/EJWQIF3XKWIBIDTBC66CCBQ7LD.json","graph_json":"https://pith.science/api/pith-number/EJWQIF3XKWIBIDTBC66CCBQ7LD/graph.json","events_json":"https://pith.science/api/pith-number/EJWQIF3XKWIBIDTBC66CCBQ7LD/events.json","paper":"https://pith.science/paper/EJWQIF3X"},"agent_actions":{"view_html":"https://pith.science/pith/EJWQIF3XKWIBIDTBC66CCBQ7LD","download_json":"https://pith.science/pith/EJWQIF3XKWIBIDTBC66CCBQ7LD.json","view_paper":"https://pith.science/paper/EJWQIF3X","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2411.05345&json=true","fetch_graph":"https://pith.science/api/pith-number/EJWQIF3XKWIBIDTBC66CCBQ7LD/graph.json","fetch_events":"https://pith.science/api/pith-number/EJWQIF3XKWIBIDTBC66CCBQ7LD/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/EJWQIF3XKWIBIDTBC66CCBQ7LD/action/timestamp_anchor","attest_storage":"https://pith.science/pith/EJWQIF3XKWIBIDTBC66CCBQ7LD/action/storage_attestation","attest_author":"https://pith.science/pith/EJWQIF3XKWIBIDTBC66CCBQ7LD/action/author_attestation","sign_citation":"https://pith.science/pith/EJWQIF3XKWIBIDTBC66CCBQ7LD/action/citation_signature","submit_replication":"https://pith.science/pith/EJWQIF3XKWIBIDTBC66CCBQ7LD/action/replication_record"}},"created_at":"2026-07-05T09:32:59.007320+00:00","updated_at":"2026-07-05T09:32:59.007320+00:00"}