{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:ZLM2KLWHBT42XV32KGJETQ4OHE","short_pith_number":"pith:ZLM2KLWH","schema_version":"1.0","canonical_sha256":"cad9a52ec70cf9abd77a519249c38e39361c7b98effc5b9af2dd0dea200d6d32","source":{"kind":"arxiv","id":"2410.06672","version":2},"attestation_state":"computed","paper":{"title":"Towards Universality: Studying Mechanistic Similarity Across Language Model Architectures","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Junxuan Wang, Qiong Tang, Wentao Shu, Xipeng Qiu, Xuyang Ge, Yunhua Zhou, Zhengfu He","submitted_at":"2024-10-09T08:28:53Z","abstract_excerpt":"The hypothesis of Universality in interpretability suggests that different neural networks may converge to implement similar algorithms on similar tasks. In this work, we investigate two mainstream architectures for language modeling, namely Transformers and Mambas, to explore the extent of their mechanistic similarity. We propose to use Sparse Autoencoders (SAEs) to isolate interpretable features from these models and show that most features are similar in these two models. We also validate the correlation between feature similarity and Universality. We then delve into the circuit-level analy"},"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":"2410.06672","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2024-10-09T08:28:53Z","cross_cats_sorted":[],"title_canon_sha256":"dcc07eb1a80600d8d95f88aa4e547c3d7836064c37046068db7049a815f2cdac","abstract_canon_sha256":"5f1e00b40947633b1641f66cf6004e848553d4a8905851eb70351502568f5a8c"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:18:31.952211Z","signature_b64":"A/AZzUdq5Zp/eYn3oNf4D9lJHwbXwMiBt5lyVEIgMalAL23QSGFDNc/QQKUwpFIceIs4QKGM5+dqWazBjAEpBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"cad9a52ec70cf9abd77a519249c38e39361c7b98effc5b9af2dd0dea200d6d32","last_reissued_at":"2026-07-05T09:18:31.951725Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:18:31.951725Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Towards Universality: Studying Mechanistic Similarity Across Language Model Architectures","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Junxuan Wang, Qiong Tang, Wentao Shu, Xipeng Qiu, Xuyang Ge, Yunhua Zhou, Zhengfu He","submitted_at":"2024-10-09T08:28:53Z","abstract_excerpt":"The hypothesis of Universality in interpretability suggests that different neural networks may converge to implement similar algorithms on similar tasks. In this work, we investigate two mainstream architectures for language modeling, namely Transformers and Mambas, to explore the extent of their mechanistic similarity. We propose to use Sparse Autoencoders (SAEs) to isolate interpretable features from these models and show that most features are similar in these two models. We also validate the correlation between feature similarity and Universality. We then delve into the circuit-level analy"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2410.06672","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/2410.06672/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":"2410.06672","created_at":"2026-07-05T09:18:31.951783+00:00"},{"alias_kind":"arxiv_version","alias_value":"2410.06672v2","created_at":"2026-07-05T09:18:31.951783+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2410.06672","created_at":"2026-07-05T09:18:31.951783+00:00"},{"alias_kind":"pith_short_12","alias_value":"ZLM2KLWHBT42","created_at":"2026-07-05T09:18:31.951783+00:00"},{"alias_kind":"pith_short_16","alias_value":"ZLM2KLWHBT42XV32","created_at":"2026-07-05T09:18:31.951783+00:00"},{"alias_kind":"pith_short_8","alias_value":"ZLM2KLWH","created_at":"2026-07-05T09:18:31.951783+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.04717","citing_title":"Auditing CoT Answer-Hijack Patches: Source-Control Certificates with Type-I Guarantees","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12770","citing_title":"WriteSAE: Sparse Autoencoders for Recurrent State","ref_index":45,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12770","citing_title":"WriteSAE: Sparse Autoencoders for Recurrent State","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12770","citing_title":"WriteSAE: Sparse Autoencoders for Recurrent State","ref_index":18,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/ZLM2KLWHBT42XV32KGJETQ4OHE","json":"https://pith.science/pith/ZLM2KLWHBT42XV32KGJETQ4OHE.json","graph_json":"https://pith.science/api/pith-number/ZLM2KLWHBT42XV32KGJETQ4OHE/graph.json","events_json":"https://pith.science/api/pith-number/ZLM2KLWHBT42XV32KGJETQ4OHE/events.json","paper":"https://pith.science/paper/ZLM2KLWH"},"agent_actions":{"view_html":"https://pith.science/pith/ZLM2KLWHBT42XV32KGJETQ4OHE","download_json":"https://pith.science/pith/ZLM2KLWHBT42XV32KGJETQ4OHE.json","view_paper":"https://pith.science/paper/ZLM2KLWH","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2410.06672&json=true","fetch_graph":"https://pith.science/api/pith-number/ZLM2KLWHBT42XV32KGJETQ4OHE/graph.json","fetch_events":"https://pith.science/api/pith-number/ZLM2KLWHBT42XV32KGJETQ4OHE/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ZLM2KLWHBT42XV32KGJETQ4OHE/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ZLM2KLWHBT42XV32KGJETQ4OHE/action/storage_attestation","attest_author":"https://pith.science/pith/ZLM2KLWHBT42XV32KGJETQ4OHE/action/author_attestation","sign_citation":"https://pith.science/pith/ZLM2KLWHBT42XV32KGJETQ4OHE/action/citation_signature","submit_replication":"https://pith.science/pith/ZLM2KLWHBT42XV32KGJETQ4OHE/action/replication_record"}},"created_at":"2026-07-05T09:18:31.951783+00:00","updated_at":"2026-07-05T09:18:31.951783+00:00"}