{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:PG4UEOEC5B6P7F6M5FKOBJXUWL","short_pith_number":"pith:PG4UEOEC","schema_version":"1.0","canonical_sha256":"79b9423882e87cff97cce954e0a6f4b2caa8063c2f36e6da5d167c47cf3d9c57","source":{"kind":"arxiv","id":"2502.17516","version":1},"attestation_state":"computed","paper":{"title":"A Survey on Mechanistic Interpretability for Multi-Modal Foundation Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Arman Zarei, Barry Menglong Yao, Hongxuan Li, Keivan Rezaei, Lifu Huang, Li Shen, Mohammad Beigi, Qin Liu, Ryan A. Rossi, Samyadeep Basu, Shilong Liu, Soheil Feizi, Sriram Balasubramanian, Varun Manjunatha, Yan Sun, Ying Shen, Yufan Zhou, Yuxiang Zhang, Zhiyang Xu, Zichao Wang, Zihao Lin","submitted_at":"2025-02-22T20:55:26Z","abstract_excerpt":"The rise of foundation models has transformed machine learning research, prompting efforts to uncover their inner workings and develop more efficient and reliable applications for better control. While significant progress has been made in interpreting Large Language Models (LLMs), multimodal foundation models (MMFMs) - such as contrastive vision-language models, generative vision-language models, and text-to-image models - pose unique interpretability challenges beyond unimodal frameworks. Despite initial studies, a substantial gap remains between the interpretability of LLMs and MMFMs. This "},"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":"2502.17516","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-02-22T20:55:26Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"3ee6c9f36bdc3d78afde30da7f4b695b05dc89941034be471c493d8b5f998985","abstract_canon_sha256":"1d18774de2ddd85ae113c31bec0434a27b8cdf293dcbebffdd65987c903fec77"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:19:11.935862Z","signature_b64":"biPv9kdtk4o/InmXPJxFP/Y0Un5/xRMnay1sn3rdj169mEI8pJhHjhxvqMqQ8jvZbDOCJZ8R+Lxt7k/aAI7bAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"79b9423882e87cff97cce954e0a6f4b2caa8063c2f36e6da5d167c47cf3d9c57","last_reissued_at":"2026-07-05T10:19:11.935357Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:19:11.935357Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"A Survey on Mechanistic Interpretability for Multi-Modal Foundation Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Arman Zarei, Barry Menglong Yao, Hongxuan Li, Keivan Rezaei, Lifu Huang, Li Shen, Mohammad Beigi, Qin Liu, Ryan A. Rossi, Samyadeep Basu, Shilong Liu, Soheil Feizi, Sriram Balasubramanian, Varun Manjunatha, Yan Sun, Ying Shen, Yufan Zhou, Yuxiang Zhang, Zhiyang Xu, Zichao Wang, Zihao Lin","submitted_at":"2025-02-22T20:55:26Z","abstract_excerpt":"The rise of foundation models has transformed machine learning research, prompting efforts to uncover their inner workings and develop more efficient and reliable applications for better control. While significant progress has been made in interpreting Large Language Models (LLMs), multimodal foundation models (MMFMs) - such as contrastive vision-language models, generative vision-language models, and text-to-image models - pose unique interpretability challenges beyond unimodal frameworks. Despite initial studies, a substantial gap remains between the interpretability of LLMs and MMFMs. This "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.17516","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/2502.17516/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":"2502.17516","created_at":"2026-07-05T10:19:11.935420+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.17516v1","created_at":"2026-07-05T10:19:11.935420+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.17516","created_at":"2026-07-05T10:19:11.935420+00:00"},{"alias_kind":"pith_short_12","alias_value":"PG4UEOEC5B6P","created_at":"2026-07-05T10:19:11.935420+00:00"},{"alias_kind":"pith_short_16","alias_value":"PG4UEOEC5B6P7F6M","created_at":"2026-07-05T10:19:11.935420+00:00"},{"alias_kind":"pith_short_8","alias_value":"PG4UEOEC","created_at":"2026-07-05T10:19:11.935420+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":11,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.18924","citing_title":"Who Wins the Conflict? Mechanistic Interpretability of Text Bias in Audio LLMs","ref_index":24,"is_internal_anchor":false},{"citing_arxiv_id":"2606.06840","citing_title":"Characterize Then Distill: Mechanistic Reasoning in Large Output Spaces","ref_index":83,"is_internal_anchor":false},{"citing_arxiv_id":"2605.23778","citing_title":"The physics of AI weather models","ref_index":26,"is_internal_anchor":false},{"citing_arxiv_id":"2505.16120","citing_title":"LLM-Powered AI Agent Systems and Their Applications in Industry","ref_index":43,"is_internal_anchor":false},{"citing_arxiv_id":"2605.22170","citing_title":"Do Factual Recall Mechanisms Carry over from Text to Speech in Multimodal Language Models?","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2509.14837","citing_title":"V-SEAM: Visual Semantic Editing and Attention Modulating for Causal Interpretability of Vision-Language Models","ref_index":25,"is_internal_anchor":false},{"citing_arxiv_id":"2601.14004","citing_title":"Locate, Steer, and Improve: A Practical Survey of Actionable Mechanistic Interpretability in Large Language Models","ref_index":187,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08809","citing_title":"SimReg: Achieving Higher Performance in the Pretraining via Embedding Similarity Regularization","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2604.10172","citing_title":"Wearable AI in the Era of Large Sensor Models","ref_index":23,"is_internal_anchor":false},{"citing_arxiv_id":"2604.16902","citing_title":"Beyond Text-Dominance: Understanding Modality Preference of Omni-modal Large Language Models","ref_index":26,"is_internal_anchor":false},{"citing_arxiv_id":"2604.17941","citing_title":"From Heads to Neurons: Causal Attribution and Steering in Multi-Task Vision-Language Models","ref_index":88,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/PG4UEOEC5B6P7F6M5FKOBJXUWL","json":"https://pith.science/pith/PG4UEOEC5B6P7F6M5FKOBJXUWL.json","graph_json":"https://pith.science/api/pith-number/PG4UEOEC5B6P7F6M5FKOBJXUWL/graph.json","events_json":"https://pith.science/api/pith-number/PG4UEOEC5B6P7F6M5FKOBJXUWL/events.json","paper":"https://pith.science/paper/PG4UEOEC"},"agent_actions":{"view_html":"https://pith.science/pith/PG4UEOEC5B6P7F6M5FKOBJXUWL","download_json":"https://pith.science/pith/PG4UEOEC5B6P7F6M5FKOBJXUWL.json","view_paper":"https://pith.science/paper/PG4UEOEC","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.17516&json=true","fetch_graph":"https://pith.science/api/pith-number/PG4UEOEC5B6P7F6M5FKOBJXUWL/graph.json","fetch_events":"https://pith.science/api/pith-number/PG4UEOEC5B6P7F6M5FKOBJXUWL/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/PG4UEOEC5B6P7F6M5FKOBJXUWL/action/timestamp_anchor","attest_storage":"https://pith.science/pith/PG4UEOEC5B6P7F6M5FKOBJXUWL/action/storage_attestation","attest_author":"https://pith.science/pith/PG4UEOEC5B6P7F6M5FKOBJXUWL/action/author_attestation","sign_citation":"https://pith.science/pith/PG4UEOEC5B6P7F6M5FKOBJXUWL/action/citation_signature","submit_replication":"https://pith.science/pith/PG4UEOEC5B6P7F6M5FKOBJXUWL/action/replication_record"}},"created_at":"2026-07-05T10:19:11.935420+00:00","updated_at":"2026-07-05T10:19:11.935420+00:00"}