{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:2MNSEPXWX4NPATR2UXQ4QSNTX7","short_pith_number":"pith:2MNSEPXW","schema_version":"1.0","canonical_sha256":"d31b223ef6bf1af04e3aa5e1c849b3bfed563bdd7fcb1a687dfd3d4957be0f13","source":{"kind":"arxiv","id":"2310.12921","version":2},"attestation_state":"computed","paper":{"title":"Vision-Language Models are Zero-Shot Reward Models for Reinforcement Learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"David Lindner, Elvis Nava, Ethan Perez, Juan Rocamonde, Victoriano Montesinos","submitted_at":"2023-10-19T17:17:06Z","abstract_excerpt":"Reinforcement learning (RL) requires either manually specifying a reward function, which is often infeasible, or learning a reward model from a large amount of human feedback, which is often very expensive. We study a more sample-efficient alternative: using pretrained vision-language models (VLMs) as zero-shot reward models (RMs) to specify tasks via natural language. We propose a natural and general approach to using VLMs as reward models, which we call VLM-RMs. We use VLM-RMs based on CLIP to train a MuJoCo humanoid to learn complex tasks without a manually specified reward function, such 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":"2310.12921","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2023-10-19T17:17:06Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"3b1916a403453d8261ec074ad619beaee32e2b5a019f3a5f89894a7a8ff6561d","abstract_canon_sha256":"d475c89ac3c1840711c23841f8011cef58d2f7921d73da5de123a40e1dde866d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:56:00.617943Z","signature_b64":"HBdaziHK/nWJXVyjo8BOZWBhx3q/2j3JyzHR1ihYXcmKsCjJpfyH8r9baEMmXWIzAx3PSS3nmWu1ebc+UfGkCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"d31b223ef6bf1af04e3aa5e1c849b3bfed563bdd7fcb1a687dfd3d4957be0f13","last_reissued_at":"2026-07-05T07:56:00.617394Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:56:00.617394Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Vision-Language Models are Zero-Shot Reward Models for Reinforcement Learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"David Lindner, Elvis Nava, Ethan Perez, Juan Rocamonde, Victoriano Montesinos","submitted_at":"2023-10-19T17:17:06Z","abstract_excerpt":"Reinforcement learning (RL) requires either manually specifying a reward function, which is often infeasible, or learning a reward model from a large amount of human feedback, which is often very expensive. We study a more sample-efficient alternative: using pretrained vision-language models (VLMs) as zero-shot reward models (RMs) to specify tasks via natural language. We propose a natural and general approach to using VLMs as reward models, which we call VLM-RMs. We use VLM-RMs based on CLIP to train a MuJoCo humanoid to learn complex tasks without a manually specified reward function, such a"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2310.12921","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/2310.12921/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":"2310.12921","created_at":"2026-07-05T07:56:00.617456+00:00"},{"alias_kind":"arxiv_version","alias_value":"2310.12921v2","created_at":"2026-07-05T07:56:00.617456+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2310.12921","created_at":"2026-07-05T07:56:00.617456+00:00"},{"alias_kind":"pith_short_12","alias_value":"2MNSEPXWX4NP","created_at":"2026-07-05T07:56:00.617456+00:00"},{"alias_kind":"pith_short_16","alias_value":"2MNSEPXWX4NPATR2","created_at":"2026-07-05T07:56:00.617456+00:00"},{"alias_kind":"pith_short_8","alias_value":"2MNSEPXW","created_at":"2026-07-05T07:56:00.617456+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":9,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.05391","citing_title":"LLM-as-a-Verifier: A General-Purpose Verification Framework","ref_index":79,"is_internal_anchor":true},{"citing_arxiv_id":"2606.23640","citing_title":"Learning Process Rewards via Success Visitation Matching for Efficient RL","ref_index":71,"is_internal_anchor":false},{"citing_arxiv_id":"2606.13675","citing_title":"Improving Robotic Generalist Policies via Flow Reversal Steering","ref_index":41,"is_internal_anchor":false},{"citing_arxiv_id":"2606.28760","citing_title":"Vision-Language Models for Deployable Social Robot Navigation: Bridging Semantic Reasoning and Low-Level Control","ref_index":26,"is_internal_anchor":false},{"citing_arxiv_id":"2606.12072","citing_title":"World Model Self-Distillation: Training World Models to Solve General Tasks","ref_index":47,"is_internal_anchor":false},{"citing_arxiv_id":"2605.17933","citing_title":"AtlasVA: Self-Evolving Visual Skill Memory for Teacher-Free VLM Agents","ref_index":32,"is_internal_anchor":false},{"citing_arxiv_id":"2605.18455","citing_title":"OrganicHAR: Towards Activity Discovery in Organic Settings for Privacy Preserving Sensors Using Efficient Video Analysis","ref_index":62,"is_internal_anchor":false},{"citing_arxiv_id":"2510.27157","citing_title":"A Survey on Generative Recommendation: Data, Model, and Tasks","ref_index":152,"is_internal_anchor":false},{"citing_arxiv_id":"2307.05973","citing_title":"VoxPoser: Composable 3D Value Maps for Robotic Manipulation with Language Models","ref_index":87,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/2MNSEPXWX4NPATR2UXQ4QSNTX7","json":"https://pith.science/pith/2MNSEPXWX4NPATR2UXQ4QSNTX7.json","graph_json":"https://pith.science/api/pith-number/2MNSEPXWX4NPATR2UXQ4QSNTX7/graph.json","events_json":"https://pith.science/api/pith-number/2MNSEPXWX4NPATR2UXQ4QSNTX7/events.json","paper":"https://pith.science/paper/2MNSEPXW"},"agent_actions":{"view_html":"https://pith.science/pith/2MNSEPXWX4NPATR2UXQ4QSNTX7","download_json":"https://pith.science/pith/2MNSEPXWX4NPATR2UXQ4QSNTX7.json","view_paper":"https://pith.science/paper/2MNSEPXW","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2310.12921&json=true","fetch_graph":"https://pith.science/api/pith-number/2MNSEPXWX4NPATR2UXQ4QSNTX7/graph.json","fetch_events":"https://pith.science/api/pith-number/2MNSEPXWX4NPATR2UXQ4QSNTX7/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/2MNSEPXWX4NPATR2UXQ4QSNTX7/action/timestamp_anchor","attest_storage":"https://pith.science/pith/2MNSEPXWX4NPATR2UXQ4QSNTX7/action/storage_attestation","attest_author":"https://pith.science/pith/2MNSEPXWX4NPATR2UXQ4QSNTX7/action/author_attestation","sign_citation":"https://pith.science/pith/2MNSEPXWX4NPATR2UXQ4QSNTX7/action/citation_signature","submit_replication":"https://pith.science/pith/2MNSEPXWX4NPATR2UXQ4QSNTX7/action/replication_record"}},"created_at":"2026-07-05T07:56:00.617456+00:00","updated_at":"2026-07-05T07:56:00.617456+00:00"}