{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:RX4XAAFVPCSEX5I6BAZZ2Q6XWA","short_pith_number":"pith:RX4XAAFV","schema_version":"1.0","canonical_sha256":"8df97000b578a44bf51e08339d43d7b03fa1487ed727eafc1d94459ac9182ad5","source":{"kind":"arxiv","id":"2501.03012","version":2},"attestation_state":"computed","paper":{"title":"Analyzing Finetuning Representation Shift for Multimodal LLMs Steering","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.CL","cs.CV"],"primary_cat":"cs.AI","authors_text":"Arnaud Dapogny, Jayneel Parekh, Matthieu Cord, Mustafa Shukor, Pegah Khayatan","submitted_at":"2025-01-06T13:37:13Z","abstract_excerpt":"Multimodal LLMs (MLLMs) have reached remarkable levels of proficiency in understanding multimodal inputs. However, understanding and interpreting the behavior of such complex models is a challenging task, not to mention the dynamic shifts that may occur during fine-tuning, or due to covariate shift between datasets. In this work, we apply concept-level analysis towards MLLM understanding. More specifically, we propose to map hidden states to interpretable visual and textual concepts. This enables us to more efficiently compare certain semantic dynamics, such as the shift from an original and f"},"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":"2501.03012","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.AI","submitted_at":"2025-01-06T13:37:13Z","cross_cats_sorted":["cs.CL","cs.CV"],"title_canon_sha256":"378f482f30cc7ec6da14520c54310ba7bcc4e841a3247743817d996caa4fda0c","abstract_canon_sha256":"601b6971fc45519d3b0fe8132001bae799e8e60160d89da9dd1150505d7b0b04"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:52:58.705135Z","signature_b64":"wXO7yibzC1/eVEb2nIUlcUuUfAbo/seQ9T23OEKzBV+2NmDdPcZUcTMOyR2+GjEPt9kjEdIGarnaKuWfKhzQBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"8df97000b578a44bf51e08339d43d7b03fa1487ed727eafc1d94459ac9182ad5","last_reissued_at":"2026-07-05T11:52:58.704587Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:52:58.704587Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Analyzing Finetuning Representation Shift for Multimodal LLMs Steering","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.CL","cs.CV"],"primary_cat":"cs.AI","authors_text":"Arnaud Dapogny, Jayneel Parekh, Matthieu Cord, Mustafa Shukor, Pegah Khayatan","submitted_at":"2025-01-06T13:37:13Z","abstract_excerpt":"Multimodal LLMs (MLLMs) have reached remarkable levels of proficiency in understanding multimodal inputs. However, understanding and interpreting the behavior of such complex models is a challenging task, not to mention the dynamic shifts that may occur during fine-tuning, or due to covariate shift between datasets. In this work, we apply concept-level analysis towards MLLM understanding. More specifically, we propose to map hidden states to interpretable visual and textual concepts. This enables us to more efficiently compare certain semantic dynamics, such as the shift from an original and f"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2501.03012","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/2501.03012/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":"2501.03012","created_at":"2026-07-05T11:52:58.704645+00:00"},{"alias_kind":"arxiv_version","alias_value":"2501.03012v2","created_at":"2026-07-05T11:52:58.704645+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2501.03012","created_at":"2026-07-05T11:52:58.704645+00:00"},{"alias_kind":"pith_short_12","alias_value":"RX4XAAFVPCSE","created_at":"2026-07-05T11:52:58.704645+00:00"},{"alias_kind":"pith_short_16","alias_value":"RX4XAAFVPCSEX5I6","created_at":"2026-07-05T11:52:58.704645+00:00"},{"alias_kind":"pith_short_8","alias_value":"RX4XAAFV","created_at":"2026-07-05T11:52:58.704645+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2507.18043","citing_title":"GrAInS: Gradient-based Attribution for Inference-Time Steering of LLMs and VLMs","ref_index":18,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/RX4XAAFVPCSEX5I6BAZZ2Q6XWA","json":"https://pith.science/pith/RX4XAAFVPCSEX5I6BAZZ2Q6XWA.json","graph_json":"https://pith.science/api/pith-number/RX4XAAFVPCSEX5I6BAZZ2Q6XWA/graph.json","events_json":"https://pith.science/api/pith-number/RX4XAAFVPCSEX5I6BAZZ2Q6XWA/events.json","paper":"https://pith.science/paper/RX4XAAFV"},"agent_actions":{"view_html":"https://pith.science/pith/RX4XAAFVPCSEX5I6BAZZ2Q6XWA","download_json":"https://pith.science/pith/RX4XAAFVPCSEX5I6BAZZ2Q6XWA.json","view_paper":"https://pith.science/paper/RX4XAAFV","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2501.03012&json=true","fetch_graph":"https://pith.science/api/pith-number/RX4XAAFVPCSEX5I6BAZZ2Q6XWA/graph.json","fetch_events":"https://pith.science/api/pith-number/RX4XAAFVPCSEX5I6BAZZ2Q6XWA/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/RX4XAAFVPCSEX5I6BAZZ2Q6XWA/action/timestamp_anchor","attest_storage":"https://pith.science/pith/RX4XAAFVPCSEX5I6BAZZ2Q6XWA/action/storage_attestation","attest_author":"https://pith.science/pith/RX4XAAFVPCSEX5I6BAZZ2Q6XWA/action/author_attestation","sign_citation":"https://pith.science/pith/RX4XAAFVPCSEX5I6BAZZ2Q6XWA/action/citation_signature","submit_replication":"https://pith.science/pith/RX4XAAFVPCSEX5I6BAZZ2Q6XWA/action/replication_record"}},"created_at":"2026-07-05T11:52:58.704645+00:00","updated_at":"2026-07-05T11:52:58.704645+00:00"}