{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:DBD6XTPGSTOZ3CBVEGMBNGASMZ","short_pith_number":"pith:DBD6XTPG","schema_version":"1.0","canonical_sha256":"1847ebcde694dd9d88352198169812665630edc037726bcd3a3e20efdd3014c3","source":{"kind":"arxiv","id":"2507.00322","version":2},"attestation_state":"computed","paper":{"title":"Failure by Interference: Language Models Make Balanced Parentheses Errors When Faulty Mechanisms Overshadow Sound Ones","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.SE"],"primary_cat":"cs.CL","authors_text":"Daking Rai, Kevin Moran, Samuel Miller, Ziyu Yao","submitted_at":"2025-06-30T23:35:19Z","abstract_excerpt":"Despite remarkable advances in coding capabilities, language models (LMs) still struggle with simple syntactic tasks such as generating balanced parentheses. In this study, we investigate the underlying mechanisms behind the persistence of these errors across LMs of varying sizes (124M-7B) to both understand and mitigate the errors. Our study reveals that LMs rely on a number of components (attention heads and FF neurons) that independently make their own predictions. While some components reliably promote correct answers across a generalized range of inputs (i.e., implementing \"sound mechanis"},"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":"2507.00322","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-06-30T23:35:19Z","cross_cats_sorted":["cs.AI","cs.SE"],"title_canon_sha256":"90d9862e5b230046b13d931f7b358aff5bae0517b46a91733301508ffb881818","abstract_canon_sha256":"7a363484e3f0184bda08ef8da08eeca490f4cc626144c06c5912eb3385b6e171"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-09T02:08:31.582388Z","signature_b64":"vK+ZWTynX8zDffC0orIF33ehYoWxUF3S+aZxXPCuME7hr4vZ1mhgqq7fW1jWKzJf1Si8b0OD59bPVDQIKmHgDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"1847ebcde694dd9d88352198169812665630edc037726bcd3a3e20efdd3014c3","last_reissued_at":"2026-06-09T02:08:31.581894Z","signature_status":"signed_v1","first_computed_at":"2026-06-09T02:08:31.581894Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Failure by Interference: Language Models Make Balanced Parentheses Errors When Faulty Mechanisms Overshadow Sound Ones","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.SE"],"primary_cat":"cs.CL","authors_text":"Daking Rai, Kevin Moran, Samuel Miller, Ziyu Yao","submitted_at":"2025-06-30T23:35:19Z","abstract_excerpt":"Despite remarkable advances in coding capabilities, language models (LMs) still struggle with simple syntactic tasks such as generating balanced parentheses. In this study, we investigate the underlying mechanisms behind the persistence of these errors across LMs of varying sizes (124M-7B) to both understand and mitigate the errors. Our study reveals that LMs rely on a number of components (attention heads and FF neurons) that independently make their own predictions. While some components reliably promote correct answers across a generalized range of inputs (i.e., implementing \"sound mechanis"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2507.00322","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/2507.00322/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":"2507.00322","created_at":"2026-06-09T02:08:31.581952+00:00"},{"alias_kind":"arxiv_version","alias_value":"2507.00322v2","created_at":"2026-06-09T02:08:31.581952+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2507.00322","created_at":"2026-06-09T02:08:31.581952+00:00"},{"alias_kind":"pith_short_12","alias_value":"DBD6XTPGSTOZ","created_at":"2026-06-09T02:08:31.581952+00:00"},{"alias_kind":"pith_short_16","alias_value":"DBD6XTPGSTOZ3CBV","created_at":"2026-06-09T02:08:31.581952+00:00"},{"alias_kind":"pith_short_8","alias_value":"DBD6XTPG","created_at":"2026-06-09T02:08:31.581952+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/DBD6XTPGSTOZ3CBVEGMBNGASMZ","json":"https://pith.science/pith/DBD6XTPGSTOZ3CBVEGMBNGASMZ.json","graph_json":"https://pith.science/api/pith-number/DBD6XTPGSTOZ3CBVEGMBNGASMZ/graph.json","events_json":"https://pith.science/api/pith-number/DBD6XTPGSTOZ3CBVEGMBNGASMZ/events.json","paper":"https://pith.science/paper/DBD6XTPG"},"agent_actions":{"view_html":"https://pith.science/pith/DBD6XTPGSTOZ3CBVEGMBNGASMZ","download_json":"https://pith.science/pith/DBD6XTPGSTOZ3CBVEGMBNGASMZ.json","view_paper":"https://pith.science/paper/DBD6XTPG","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2507.00322&json=true","fetch_graph":"https://pith.science/api/pith-number/DBD6XTPGSTOZ3CBVEGMBNGASMZ/graph.json","fetch_events":"https://pith.science/api/pith-number/DBD6XTPGSTOZ3CBVEGMBNGASMZ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/DBD6XTPGSTOZ3CBVEGMBNGASMZ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/DBD6XTPGSTOZ3CBVEGMBNGASMZ/action/storage_attestation","attest_author":"https://pith.science/pith/DBD6XTPGSTOZ3CBVEGMBNGASMZ/action/author_attestation","sign_citation":"https://pith.science/pith/DBD6XTPGSTOZ3CBVEGMBNGASMZ/action/citation_signature","submit_replication":"https://pith.science/pith/DBD6XTPGSTOZ3CBVEGMBNGASMZ/action/replication_record"}},"created_at":"2026-06-09T02:08:31.581952+00:00","updated_at":"2026-06-09T02:08:31.581952+00:00"}