{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:ESVGXQOP5MORKFBC5K6EIBQQTJ","short_pith_number":"pith:ESVGXQOP","schema_version":"1.0","canonical_sha256":"24aa6bc1cfeb1d151422eabc4406109a7143e6180f1efe5a1c595852248a46ca","source":{"kind":"arxiv","id":"2503.07639","version":1},"attestation_state":"computed","paper":{"title":"Mixture of Experts Made Intrinsically Interpretable","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Adel Bibi, Ashkan Khakzar, Christian Schroeder de Witt, Constantin Venhoff, Philip Torr, Puneet K. Dokania, Xingyi Yang","submitted_at":"2025-03-05T17:40:54Z","abstract_excerpt":"Neurons in large language models often exhibit \\emph{polysemanticity}, simultaneously encoding multiple unrelated concepts and obscuring interpretability. Instead of relying on post-hoc methods, we present \\textbf{MoE-X}, a Mixture-of-Experts (MoE) language model designed to be \\emph{intrinsically} interpretable. Our approach is motivated by the observation that, in language models, wider networks with sparse activations are more likely to capture interpretable factors. However, directly training such large sparse networks is computationally prohibitive. MoE architectures offer a scalable alte"},"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":"2503.07639","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-03-05T17:40:54Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"0c4201063be64ce228b857c6572db553f2635f8b7f3a14ca07cf5f195371c5fd","abstract_canon_sha256":"e34deb6ede60feb05e3f0857c4a0dea15ce7147773a2b1f61667d416786c37bc"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:28:40.573958Z","signature_b64":"pb9PKrDn1/gxV5IdcKaLDs6p/MfNoyANWNKDXii7j1OVPrj9IVHEneZkxrYjir51BjjeXXTxJmougtg6mB8dDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"24aa6bc1cfeb1d151422eabc4406109a7143e6180f1efe5a1c595852248a46ca","last_reissued_at":"2026-07-05T10:28:40.573145Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:28:40.573145Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Mixture of Experts Made Intrinsically Interpretable","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Adel Bibi, Ashkan Khakzar, Christian Schroeder de Witt, Constantin Venhoff, Philip Torr, Puneet K. Dokania, Xingyi Yang","submitted_at":"2025-03-05T17:40:54Z","abstract_excerpt":"Neurons in large language models often exhibit \\emph{polysemanticity}, simultaneously encoding multiple unrelated concepts and obscuring interpretability. Instead of relying on post-hoc methods, we present \\textbf{MoE-X}, a Mixture-of-Experts (MoE) language model designed to be \\emph{intrinsically} interpretable. Our approach is motivated by the observation that, in language models, wider networks with sparse activations are more likely to capture interpretable factors. However, directly training such large sparse networks is computationally prohibitive. MoE architectures offer a scalable alte"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2503.07639","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/2503.07639/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":"2503.07639","created_at":"2026-07-05T10:28:40.573225+00:00"},{"alias_kind":"arxiv_version","alias_value":"2503.07639v1","created_at":"2026-07-05T10:28:40.573225+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2503.07639","created_at":"2026-07-05T10:28:40.573225+00:00"},{"alias_kind":"pith_short_12","alias_value":"ESVGXQOP5MOR","created_at":"2026-07-05T10:28:40.573225+00:00"},{"alias_kind":"pith_short_16","alias_value":"ESVGXQOP5MORKFBC","created_at":"2026-07-05T10:28:40.573225+00:00"},{"alias_kind":"pith_short_8","alias_value":"ESVGXQOP","created_at":"2026-07-05T10:28:40.573225+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.02946","citing_title":"RouteHijack: Routing-Aware Attack on Mixture-of-Experts LLMs","ref_index":78,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/ESVGXQOP5MORKFBC5K6EIBQQTJ","json":"https://pith.science/pith/ESVGXQOP5MORKFBC5K6EIBQQTJ.json","graph_json":"https://pith.science/api/pith-number/ESVGXQOP5MORKFBC5K6EIBQQTJ/graph.json","events_json":"https://pith.science/api/pith-number/ESVGXQOP5MORKFBC5K6EIBQQTJ/events.json","paper":"https://pith.science/paper/ESVGXQOP"},"agent_actions":{"view_html":"https://pith.science/pith/ESVGXQOP5MORKFBC5K6EIBQQTJ","download_json":"https://pith.science/pith/ESVGXQOP5MORKFBC5K6EIBQQTJ.json","view_paper":"https://pith.science/paper/ESVGXQOP","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2503.07639&json=true","fetch_graph":"https://pith.science/api/pith-number/ESVGXQOP5MORKFBC5K6EIBQQTJ/graph.json","fetch_events":"https://pith.science/api/pith-number/ESVGXQOP5MORKFBC5K6EIBQQTJ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ESVGXQOP5MORKFBC5K6EIBQQTJ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ESVGXQOP5MORKFBC5K6EIBQQTJ/action/storage_attestation","attest_author":"https://pith.science/pith/ESVGXQOP5MORKFBC5K6EIBQQTJ/action/author_attestation","sign_citation":"https://pith.science/pith/ESVGXQOP5MORKFBC5K6EIBQQTJ/action/citation_signature","submit_replication":"https://pith.science/pith/ESVGXQOP5MORKFBC5K6EIBQQTJ/action/replication_record"}},"created_at":"2026-07-05T10:28:40.573225+00:00","updated_at":"2026-07-05T10:28:40.573225+00:00"}