{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:OIDLNEQUQSYASWNMAJOEQCFEAY","short_pith_number":"pith:OIDLNEQU","schema_version":"1.0","canonical_sha256":"7206b6921484b00959ac025c4808a40601c99769ad648656822f3e19358c5299","source":{"kind":"arxiv","id":"2402.11917","version":3},"attestation_state":"computed","paper":{"title":"A Mechanistic Analysis of a Transformer Trained on a Symbolic Multi-Step Reasoning Task","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Abhay Sheshadri, Christian Bartelt, Jannik Brinkmann, Paul Swoboda, Victor Levoso","submitted_at":"2024-02-19T08:04:25Z","abstract_excerpt":"Transformers demonstrate impressive performance on a range of reasoning benchmarks. To evaluate the degree to which these abilities are a result of actual reasoning, existing work has focused on developing sophisticated benchmarks for behavioral studies. However, these studies do not provide insights into the internal mechanisms driving the observed capabilities. To improve our understanding of the internal mechanisms of transformers, we present a comprehensive mechanistic analysis of a transformer trained on a synthetic reasoning task. We identify a set of interpretable mechanisms the model u"},"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":"2402.11917","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-02-19T08:04:25Z","cross_cats_sorted":[],"title_canon_sha256":"dfa7aae9f9dcc46267db92e41ce01cea4a43801fcb909625ef37501af5720bd5","abstract_canon_sha256":"0b5b4bbca13cb20159a51eecb5f34fd30666e3be122820eb5598b484ea0ea63e"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:38:06.297466Z","signature_b64":"GC6FH2FwgDZBcHUG+4ie8tNbISYyN9S6gnF4gC9NdqQPUFe6/mRK2kAKqWoe6vY5T50ScHdvo27z4b6gZSffBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"7206b6921484b00959ac025c4808a40601c99769ad648656822f3e19358c5299","last_reissued_at":"2026-07-05T08:38:06.296834Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:38:06.296834Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"A Mechanistic Analysis of a Transformer Trained on a Symbolic Multi-Step Reasoning Task","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Abhay Sheshadri, Christian Bartelt, Jannik Brinkmann, Paul Swoboda, Victor Levoso","submitted_at":"2024-02-19T08:04:25Z","abstract_excerpt":"Transformers demonstrate impressive performance on a range of reasoning benchmarks. To evaluate the degree to which these abilities are a result of actual reasoning, existing work has focused on developing sophisticated benchmarks for behavioral studies. However, these studies do not provide insights into the internal mechanisms driving the observed capabilities. To improve our understanding of the internal mechanisms of transformers, we present a comprehensive mechanistic analysis of a transformer trained on a synthetic reasoning task. We identify a set of interpretable mechanisms the model u"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2402.11917","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/2402.11917/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":"2402.11917","created_at":"2026-07-05T08:38:06.296909+00:00"},{"alias_kind":"arxiv_version","alias_value":"2402.11917v3","created_at":"2026-07-05T08:38:06.296909+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2402.11917","created_at":"2026-07-05T08:38:06.296909+00:00"},{"alias_kind":"pith_short_12","alias_value":"OIDLNEQUQSYA","created_at":"2026-07-05T08:38:06.296909+00:00"},{"alias_kind":"pith_short_16","alias_value":"OIDLNEQUQSYASWNM","created_at":"2026-07-05T08:38:06.296909+00:00"},{"alias_kind":"pith_short_8","alias_value":"OIDLNEQU","created_at":"2026-07-05T08:38:06.296909+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.07316","citing_title":"Mechanistic Interpretability for Neural Networks: Circuits, Sparse Features and Symbolic Reasoning","ref_index":27,"is_internal_anchor":true},{"citing_arxiv_id":"2606.03685","citing_title":"A Close Look At World Model Recovery In Supervised Fine-Tuned LLM Planners","ref_index":1,"is_internal_anchor":false},{"citing_arxiv_id":"2606.00183","citing_title":"Agentic Transformers Provably Learn to Search via Reinforcement Learning","ref_index":7,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/OIDLNEQUQSYASWNMAJOEQCFEAY","json":"https://pith.science/pith/OIDLNEQUQSYASWNMAJOEQCFEAY.json","graph_json":"https://pith.science/api/pith-number/OIDLNEQUQSYASWNMAJOEQCFEAY/graph.json","events_json":"https://pith.science/api/pith-number/OIDLNEQUQSYASWNMAJOEQCFEAY/events.json","paper":"https://pith.science/paper/OIDLNEQU"},"agent_actions":{"view_html":"https://pith.science/pith/OIDLNEQUQSYASWNMAJOEQCFEAY","download_json":"https://pith.science/pith/OIDLNEQUQSYASWNMAJOEQCFEAY.json","view_paper":"https://pith.science/paper/OIDLNEQU","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2402.11917&json=true","fetch_graph":"https://pith.science/api/pith-number/OIDLNEQUQSYASWNMAJOEQCFEAY/graph.json","fetch_events":"https://pith.science/api/pith-number/OIDLNEQUQSYASWNMAJOEQCFEAY/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/OIDLNEQUQSYASWNMAJOEQCFEAY/action/timestamp_anchor","attest_storage":"https://pith.science/pith/OIDLNEQUQSYASWNMAJOEQCFEAY/action/storage_attestation","attest_author":"https://pith.science/pith/OIDLNEQUQSYASWNMAJOEQCFEAY/action/author_attestation","sign_citation":"https://pith.science/pith/OIDLNEQUQSYASWNMAJOEQCFEAY/action/citation_signature","submit_replication":"https://pith.science/pith/OIDLNEQUQSYASWNMAJOEQCFEAY/action/replication_record"}},"created_at":"2026-07-05T08:38:06.296909+00:00","updated_at":"2026-07-05T08:38:06.296909+00:00"}