{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:PGFZW4RPFHXBPN6JMCKQLGLASW","short_pith_number":"pith:PGFZW4RP","schema_version":"1.0","canonical_sha256":"798b9b722f29ee17b7c9609505996095af9989fc13639dc18d0e75a27808b8c6","source":{"kind":"arxiv","id":"2507.01368","version":1},"attestation_state":"computed","paper":{"title":"Activation Reward Models for Few-Shot Model Alignment","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CV","authors_text":"Assaf Arbelle, Brandon Huang, Chancharik Mitra, Deva Ramanan, Gautam Rajendrakumar Gare, Leonid Karlinsky, Roei Herzig, Rogerio Feris, Tianning Chai, Trevor Darrell, Zhiqiu Lin","submitted_at":"2025-07-02T05:10:29Z","abstract_excerpt":"Aligning Large Language Models (LLMs) and Large Multimodal Models (LMMs) to human preferences is a central challenge in improving the quality of the models' generative outputs for real-world applications. A common approach is to use reward modeling to encode preferences, enabling alignment via post-training using reinforcement learning. However, traditional reward modeling is not easily adaptable to new preferences because it requires a separate reward model, commonly trained on large preference datasets. To address this, we introduce Activation Reward Models (Activation RMs) -- a novel few-sh"},"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":"2507.01368","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2025-07-02T05:10:29Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"da16ce701eb52f1032e30af4dfcab263ad30765c8acffd5672f4030ec3d9b15d","abstract_canon_sha256":"196e0579dcd3ffe8c3b495e0f0653d56a2f5c4afa1312aa3bfdbfe37338f843b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:30:47.592435Z","signature_b64":"MAPQlGz7kS3yh0YyQ2EVZCYbh3F9rONc7xef3LvCFmZwCA6fw2Uj553LUc+tMZ6KOAvbBXDIGjq6bZjZyOjdCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"798b9b722f29ee17b7c9609505996095af9989fc13639dc18d0e75a27808b8c6","last_reissued_at":"2026-07-05T11:30:47.591926Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:30:47.591926Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Activation Reward Models for Few-Shot Model Alignment","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CV","authors_text":"Assaf Arbelle, Brandon Huang, Chancharik Mitra, Deva Ramanan, Gautam Rajendrakumar Gare, Leonid Karlinsky, Roei Herzig, Rogerio Feris, Tianning Chai, Trevor Darrell, Zhiqiu Lin","submitted_at":"2025-07-02T05:10:29Z","abstract_excerpt":"Aligning Large Language Models (LLMs) and Large Multimodal Models (LMMs) to human preferences is a central challenge in improving the quality of the models' generative outputs for real-world applications. A common approach is to use reward modeling to encode preferences, enabling alignment via post-training using reinforcement learning. However, traditional reward modeling is not easily adaptable to new preferences because it requires a separate reward model, commonly trained on large preference datasets. To address this, we introduce Activation Reward Models (Activation RMs) -- a novel few-sh"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2507.01368","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/2507.01368/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":"2507.01368","created_at":"2026-07-05T11:30:47.591987+00:00"},{"alias_kind":"arxiv_version","alias_value":"2507.01368v1","created_at":"2026-07-05T11:30:47.591987+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2507.01368","created_at":"2026-07-05T11:30:47.591987+00:00"},{"alias_kind":"pith_short_12","alias_value":"PGFZW4RPFHXB","created_at":"2026-07-05T11:30:47.591987+00:00"},{"alias_kind":"pith_short_16","alias_value":"PGFZW4RPFHXBPN6J","created_at":"2026-07-05T11:30:47.591987+00:00"},{"alias_kind":"pith_short_8","alias_value":"PGFZW4RP","created_at":"2026-07-05T11:30:47.591987+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2604.21718","citing_title":"Building a Precise Video Language with Human-AI Oversight","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2604.13602","citing_title":"Reward Hacking in the Era of Large Models: Mechanisms, Emergent Misalignment, Challenges","ref_index":222,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/PGFZW4RPFHXBPN6JMCKQLGLASW","json":"https://pith.science/pith/PGFZW4RPFHXBPN6JMCKQLGLASW.json","graph_json":"https://pith.science/api/pith-number/PGFZW4RPFHXBPN6JMCKQLGLASW/graph.json","events_json":"https://pith.science/api/pith-number/PGFZW4RPFHXBPN6JMCKQLGLASW/events.json","paper":"https://pith.science/paper/PGFZW4RP"},"agent_actions":{"view_html":"https://pith.science/pith/PGFZW4RPFHXBPN6JMCKQLGLASW","download_json":"https://pith.science/pith/PGFZW4RPFHXBPN6JMCKQLGLASW.json","view_paper":"https://pith.science/paper/PGFZW4RP","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2507.01368&json=true","fetch_graph":"https://pith.science/api/pith-number/PGFZW4RPFHXBPN6JMCKQLGLASW/graph.json","fetch_events":"https://pith.science/api/pith-number/PGFZW4RPFHXBPN6JMCKQLGLASW/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/PGFZW4RPFHXBPN6JMCKQLGLASW/action/timestamp_anchor","attest_storage":"https://pith.science/pith/PGFZW4RPFHXBPN6JMCKQLGLASW/action/storage_attestation","attest_author":"https://pith.science/pith/PGFZW4RPFHXBPN6JMCKQLGLASW/action/author_attestation","sign_citation":"https://pith.science/pith/PGFZW4RPFHXBPN6JMCKQLGLASW/action/citation_signature","submit_replication":"https://pith.science/pith/PGFZW4RPFHXBPN6JMCKQLGLASW/action/replication_record"}},"created_at":"2026-07-05T11:30:47.591987+00:00","updated_at":"2026-07-05T11:30:47.591987+00:00"}