{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:BLJGFJXH5O3AJGF7BBYJ5FIG4G","short_pith_number":"pith:BLJGFJXH","schema_version":"1.0","canonical_sha256":"0ad262a6e7ebb60498bf08709e9506e1913e5b76a1e391a210c681850df3b499","source":{"kind":"arxiv","id":"2602.11910","version":2},"attestation_state":"computed","paper":{"title":"TADA! Tuning Audio Diffusion Models through Activation Steering","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.SD","authors_text":"Kamil Deja, Katarzyna Zaleska, {\\L}ukasz Staniszewski, Mateusz Modrzejewski","submitted_at":"2026-02-12T13:07:14Z","abstract_excerpt":"Audio diffusion models can synthesize high-fidelity music from text, yet achieving fine-grained control over specific musical attributes remains challenging, as their internal mechanisms for representing high-level concepts are poorly understood. In this work, we use activation patching to demonstrate that recent audio diffusion architectures exhibit a semantic bottleneck, where a small, shared subset of consecutive attention layers controls distinct musical concepts, such as the presence of specific instruments, vocals, or genres. Building on this, we systematically evaluate a broad spectrum "},"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":"2602.11910","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.SD","submitted_at":"2026-02-12T13:07:14Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"1e94755ee861a9e5d638bc9158d920963158828edd1c8ec1b9a8e59749aec250","abstract_canon_sha256":"c1f8d75468a1cda59abeb68ed05bc886cc2796cb376b2bbc2bcd23379f01fab9"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T01:05:09.040893Z","signature_b64":"6lO7GI0Uo+auWvKFIh+rEBZlDxz90ImYOZ02VTJWF62md65sKtn9lys0Qob48U+me3knBgzK1uNvgZtPy9jRBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"0ad262a6e7ebb60498bf08709e9506e1913e5b76a1e391a210c681850df3b499","last_reissued_at":"2026-05-20T01:05:09.040016Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T01:05:09.040016Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"TADA! Tuning Audio Diffusion Models through Activation Steering","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.SD","authors_text":"Kamil Deja, Katarzyna Zaleska, {\\L}ukasz Staniszewski, Mateusz Modrzejewski","submitted_at":"2026-02-12T13:07:14Z","abstract_excerpt":"Audio diffusion models can synthesize high-fidelity music from text, yet achieving fine-grained control over specific musical attributes remains challenging, as their internal mechanisms for representing high-level concepts are poorly understood. In this work, we use activation patching to demonstrate that recent audio diffusion architectures exhibit a semantic bottleneck, where a small, shared subset of consecutive attention layers controls distinct musical concepts, such as the presence of specific instruments, vocals, or genres. Building on this, we systematically evaluate a broad spectrum "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2602.11910","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/2602.11910/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":"2602.11910","created_at":"2026-05-20T01:05:09.040140+00:00"},{"alias_kind":"arxiv_version","alias_value":"2602.11910v2","created_at":"2026-05-20T01:05:09.040140+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2602.11910","created_at":"2026-05-20T01:05:09.040140+00:00"},{"alias_kind":"pith_short_12","alias_value":"BLJGFJXH5O3A","created_at":"2026-05-20T01:05:09.040140+00:00"},{"alias_kind":"pith_short_16","alias_value":"BLJGFJXH5O3AJGF7","created_at":"2026-05-20T01:05:09.040140+00:00"},{"alias_kind":"pith_short_8","alias_value":"BLJGFJXH","created_at":"2026-05-20T01:05:09.040140+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/BLJGFJXH5O3AJGF7BBYJ5FIG4G","json":"https://pith.science/pith/BLJGFJXH5O3AJGF7BBYJ5FIG4G.json","graph_json":"https://pith.science/api/pith-number/BLJGFJXH5O3AJGF7BBYJ5FIG4G/graph.json","events_json":"https://pith.science/api/pith-number/BLJGFJXH5O3AJGF7BBYJ5FIG4G/events.json","paper":"https://pith.science/paper/BLJGFJXH"},"agent_actions":{"view_html":"https://pith.science/pith/BLJGFJXH5O3AJGF7BBYJ5FIG4G","download_json":"https://pith.science/pith/BLJGFJXH5O3AJGF7BBYJ5FIG4G.json","view_paper":"https://pith.science/paper/BLJGFJXH","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2602.11910&json=true","fetch_graph":"https://pith.science/api/pith-number/BLJGFJXH5O3AJGF7BBYJ5FIG4G/graph.json","fetch_events":"https://pith.science/api/pith-number/BLJGFJXH5O3AJGF7BBYJ5FIG4G/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/BLJGFJXH5O3AJGF7BBYJ5FIG4G/action/timestamp_anchor","attest_storage":"https://pith.science/pith/BLJGFJXH5O3AJGF7BBYJ5FIG4G/action/storage_attestation","attest_author":"https://pith.science/pith/BLJGFJXH5O3AJGF7BBYJ5FIG4G/action/author_attestation","sign_citation":"https://pith.science/pith/BLJGFJXH5O3AJGF7BBYJ5FIG4G/action/citation_signature","submit_replication":"https://pith.science/pith/BLJGFJXH5O3AJGF7BBYJ5FIG4G/action/replication_record"}},"created_at":"2026-05-20T01:05:09.040140+00:00","updated_at":"2026-05-20T01:05:09.040140+00:00"}