{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:G627CSU7AXN7RRDW5FEFTGPXDS","short_pith_number":"pith:G627CSU7","schema_version":"1.0","canonical_sha256":"37b5f14a9f05dbf8c476e9485999f71cb73df229a2da7419e8619c9d1bf163cb","source":{"kind":"arxiv","id":"2309.05950","version":5},"attestation_state":"computed","paper":{"title":"Language Models as Black-Box Optimizers for Vision-Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CV","cs.LG","cs.MM"],"primary_cat":"cs.CL","authors_text":"Deepak Pathak, Deva Ramanan, Ryan Lee, Samuel Yu, Shihong Liu, Tiffany Ling, Zhiqiu Lin","submitted_at":"2023-09-12T04:03:41Z","abstract_excerpt":"Vision-language models (VLMs) pre-trained on web-scale datasets have demonstrated remarkable capabilities on downstream tasks when fine-tuned with minimal data. However, many VLMs rely on proprietary data and are not open-source, which restricts the use of white-box approaches for fine-tuning. As such, we aim to develop a black-box approach to optimize VLMs through natural language prompts, thereby avoiding the need to access model parameters, feature embeddings, or even output logits. We propose employing chat-based LLMs to search for the best text prompt for VLMs. Specifically, we adopt an a"},"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":"2309.05950","kind":"arxiv","version":5},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2023-09-12T04:03:41Z","cross_cats_sorted":["cs.CV","cs.LG","cs.MM"],"title_canon_sha256":"f2b89275b72d6dae89e988ebd0fbeebf9b68cbfc6ada1322c3c8267329022943","abstract_canon_sha256":"f0ff155fe5828ae9280e77a74ea9edc5b2a92e878c7a2d37f4715f3f3984be7f"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:18:44.112957Z","signature_b64":"7DOfTQsSumxdZcimtXxR8lCzw+/EySmYj3JyRU8+Y3YE8fW5Ec3khWCkcZ4y+mwKm9BEm4+BSvQ8cn6QIhvODw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"37b5f14a9f05dbf8c476e9485999f71cb73df229a2da7419e8619c9d1bf163cb","last_reissued_at":"2026-07-05T08:18:44.112431Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:18:44.112431Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Language Models as Black-Box Optimizers for Vision-Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CV","cs.LG","cs.MM"],"primary_cat":"cs.CL","authors_text":"Deepak Pathak, Deva Ramanan, Ryan Lee, Samuel Yu, Shihong Liu, Tiffany Ling, Zhiqiu Lin","submitted_at":"2023-09-12T04:03:41Z","abstract_excerpt":"Vision-language models (VLMs) pre-trained on web-scale datasets have demonstrated remarkable capabilities on downstream tasks when fine-tuned with minimal data. However, many VLMs rely on proprietary data and are not open-source, which restricts the use of white-box approaches for fine-tuning. As such, we aim to develop a black-box approach to optimize VLMs through natural language prompts, thereby avoiding the need to access model parameters, feature embeddings, or even output logits. We propose employing chat-based LLMs to search for the best text prompt for VLMs. Specifically, we adopt an a"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2309.05950","kind":"arxiv","version":5},"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/2309.05950/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":"2309.05950","created_at":"2026-07-05T08:18:44.112488+00:00"},{"alias_kind":"arxiv_version","alias_value":"2309.05950v5","created_at":"2026-07-05T08:18:44.112488+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2309.05950","created_at":"2026-07-05T08:18:44.112488+00:00"},{"alias_kind":"pith_short_12","alias_value":"G627CSU7AXN7","created_at":"2026-07-05T08:18:44.112488+00:00"},{"alias_kind":"pith_short_16","alias_value":"G627CSU7AXN7RRDW","created_at":"2026-07-05T08:18:44.112488+00:00"},{"alias_kind":"pith_short_8","alias_value":"G627CSU7","created_at":"2026-07-05T08:18:44.112488+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2604.21718","citing_title":"Building a Precise Video Language with Human-AI Oversight","ref_index":37,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/G627CSU7AXN7RRDW5FEFTGPXDS","json":"https://pith.science/pith/G627CSU7AXN7RRDW5FEFTGPXDS.json","graph_json":"https://pith.science/api/pith-number/G627CSU7AXN7RRDW5FEFTGPXDS/graph.json","events_json":"https://pith.science/api/pith-number/G627CSU7AXN7RRDW5FEFTGPXDS/events.json","paper":"https://pith.science/paper/G627CSU7"},"agent_actions":{"view_html":"https://pith.science/pith/G627CSU7AXN7RRDW5FEFTGPXDS","download_json":"https://pith.science/pith/G627CSU7AXN7RRDW5FEFTGPXDS.json","view_paper":"https://pith.science/paper/G627CSU7","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2309.05950&json=true","fetch_graph":"https://pith.science/api/pith-number/G627CSU7AXN7RRDW5FEFTGPXDS/graph.json","fetch_events":"https://pith.science/api/pith-number/G627CSU7AXN7RRDW5FEFTGPXDS/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/G627CSU7AXN7RRDW5FEFTGPXDS/action/timestamp_anchor","attest_storage":"https://pith.science/pith/G627CSU7AXN7RRDW5FEFTGPXDS/action/storage_attestation","attest_author":"https://pith.science/pith/G627CSU7AXN7RRDW5FEFTGPXDS/action/author_attestation","sign_citation":"https://pith.science/pith/G627CSU7AXN7RRDW5FEFTGPXDS/action/citation_signature","submit_replication":"https://pith.science/pith/G627CSU7AXN7RRDW5FEFTGPXDS/action/replication_record"}},"created_at":"2026-07-05T08:18:44.112488+00:00","updated_at":"2026-07-05T08:18:44.112488+00:00"}