{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:AUD7NZHTKA4BUIBMSOAKCGACMI","short_pith_number":"pith:AUD7NZHT","schema_version":"1.0","canonical_sha256":"0507f6e4f350381a202c9380a11802622f39a57691af382533e3fe71d75d44a6","source":{"kind":"arxiv","id":"2507.12379","version":1},"attestation_state":"computed","paper":{"title":"Probing for Arithmetic Errors in Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Alessandro Stolfo, Mrinmaya Sachan, Yucheng Sun","submitted_at":"2025-07-16T16:27:50Z","abstract_excerpt":"We investigate whether internal activations in language models can be used to detect arithmetic errors. Starting with a controlled setting of 3-digit addition, we show that simple probes can accurately decode both the model's predicted output and the correct answer from hidden states, regardless of whether the model's output is correct. Building on this, we train lightweight error detectors that predict model correctness with over 90% accuracy. We then extend our analysis to structured chain-of-thought traces on addition-only GSM8K problems and find that probes trained on simple arithmetic gen"},"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":"2507.12379","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-07-16T16:27:50Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"f9eeb807c4141f66463fa06a8d88b025ceddcafaf31ecf76bbf7ebc8ff0fba4d","abstract_canon_sha256":"a9b60b3a132c1fdc2a6a04e1150ab79b9ed8936a36bee3fb469614d9d494628e"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:38:19.280331Z","signature_b64":"r2Jnf89WaVDttyGikeu4+TPSV8qhkI2qvPyCSOr+AIVJJZ/OEZtWfqUWdjlqNAUa47cMXh+pfPX/4IS7HoxlDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"0507f6e4f350381a202c9380a11802622f39a57691af382533e3fe71d75d44a6","last_reissued_at":"2026-07-05T11:38:19.279772Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:38:19.279772Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Probing for Arithmetic Errors in Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Alessandro Stolfo, Mrinmaya Sachan, Yucheng Sun","submitted_at":"2025-07-16T16:27:50Z","abstract_excerpt":"We investigate whether internal activations in language models can be used to detect arithmetic errors. Starting with a controlled setting of 3-digit addition, we show that simple probes can accurately decode both the model's predicted output and the correct answer from hidden states, regardless of whether the model's output is correct. Building on this, we train lightweight error detectors that predict model correctness with over 90% accuracy. We then extend our analysis to structured chain-of-thought traces on addition-only GSM8K problems and find that probes trained on simple arithmetic gen"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2507.12379","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/2507.12379/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":"2507.12379","created_at":"2026-07-05T11:38:19.279833+00:00"},{"alias_kind":"arxiv_version","alias_value":"2507.12379v1","created_at":"2026-07-05T11:38:19.279833+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2507.12379","created_at":"2026-07-05T11:38:19.279833+00:00"},{"alias_kind":"pith_short_12","alias_value":"AUD7NZHTKA4B","created_at":"2026-07-05T11:38:19.279833+00:00"},{"alias_kind":"pith_short_16","alias_value":"AUD7NZHTKA4BUIBM","created_at":"2026-07-05T11:38:19.279833+00:00"},{"alias_kind":"pith_short_8","alias_value":"AUD7NZHT","created_at":"2026-07-05T11:38:19.279833+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/AUD7NZHTKA4BUIBMSOAKCGACMI","json":"https://pith.science/pith/AUD7NZHTKA4BUIBMSOAKCGACMI.json","graph_json":"https://pith.science/api/pith-number/AUD7NZHTKA4BUIBMSOAKCGACMI/graph.json","events_json":"https://pith.science/api/pith-number/AUD7NZHTKA4BUIBMSOAKCGACMI/events.json","paper":"https://pith.science/paper/AUD7NZHT"},"agent_actions":{"view_html":"https://pith.science/pith/AUD7NZHTKA4BUIBMSOAKCGACMI","download_json":"https://pith.science/pith/AUD7NZHTKA4BUIBMSOAKCGACMI.json","view_paper":"https://pith.science/paper/AUD7NZHT","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2507.12379&json=true","fetch_graph":"https://pith.science/api/pith-number/AUD7NZHTKA4BUIBMSOAKCGACMI/graph.json","fetch_events":"https://pith.science/api/pith-number/AUD7NZHTKA4BUIBMSOAKCGACMI/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/AUD7NZHTKA4BUIBMSOAKCGACMI/action/timestamp_anchor","attest_storage":"https://pith.science/pith/AUD7NZHTKA4BUIBMSOAKCGACMI/action/storage_attestation","attest_author":"https://pith.science/pith/AUD7NZHTKA4BUIBMSOAKCGACMI/action/author_attestation","sign_citation":"https://pith.science/pith/AUD7NZHTKA4BUIBMSOAKCGACMI/action/citation_signature","submit_replication":"https://pith.science/pith/AUD7NZHTKA4BUIBMSOAKCGACMI/action/replication_record"}},"created_at":"2026-07-05T11:38:19.279833+00:00","updated_at":"2026-07-05T11:38:19.279833+00:00"}