{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2023:ROU4N4IRYZ3HV3M7HKTK3YY643","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"2c26c5b752bc8facff7879651ec9d50c096133e39800c4894b64e41096fcbb67","cross_cats_sorted":["cs.AI","cs.CV"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2023-02-23T17:34:53Z","title_canon_sha256":"b9f9d0a75704678b4acde8885bfbea84f3cb70eca4a6315ff07a9f7d2fb3b1f0"},"schema_version":"1.0","source":{"id":"2302.12192","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2302.12192","created_at":"2026-05-17T23:39:21Z"},{"alias_kind":"arxiv_version","alias_value":"2302.12192v1","created_at":"2026-05-17T23:39:21Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2302.12192","created_at":"2026-05-17T23:39:21Z"},{"alias_kind":"pith_short_12","alias_value":"ROU4N4IRYZ3H","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"ROU4N4IRYZ3HV3M7","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"ROU4N4IR","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:2dd1ca0a3fd9b18f4132f26591c1606d4cff3ac4add15686da72a23d57dcfce0","target":"graph","created_at":"2026-05-17T23:39:21Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"Our method generates objects with specified colors, counts and backgrounds more accurately than the pre-trained model."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"Human feedback on image-text alignment is consistent enough to be captured by a learned reward function that generalizes to new prompts and does not introduce unintended biases during fine-tuning."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"A three-stage fine-tuning process uses human ratings to train a reward model and then improves text-to-image alignment by maximizing reward-weighted likelihood."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Fine-tuning text-to-image models with human feedback improves accuracy on prompts specifying colors, counts, and backgrounds."}],"snapshot_sha256":"6cd0b1bf33b50df1bbf267ede71a4204cbb4f4fe5e7600cee531f564453ec6e1"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"paper":{"abstract_excerpt":"Deep generative models have shown impressive results in text-to-image synthesis. However, current text-to-image models often generate images that are inadequately aligned with text prompts. We propose a fine-tuning method for aligning such models using human feedback, comprising three stages. First, we collect human feedback assessing model output alignment from a set of diverse text prompts. We then use the human-labeled image-text dataset to train a reward function that predicts human feedback. Lastly, the text-to-image model is fine-tuned by maximizing reward-weighted likelihood to improve ","authors_text":"Craig Boutilier, Hao Liu, Kimin Lee, Mohammad Ghavamzadeh, Moonkyung Ryu, Olivia Watkins, Pieter Abbeel, Shixiang Shane Gu, Yuqing Du","cross_cats":["cs.AI","cs.CV"],"headline":"Fine-tuning text-to-image models with human feedback improves accuracy on prompts specifying colors, counts, and backgrounds.","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2023-02-23T17:34:53Z","title":"Aligning Text-to-Image Models using Human Feedback"},"references":{"count":26,"internal_anchors":17,"resolved_work":26,"sample":[{"cited_arxiv_id":"2112.00861","doi":"","is_internal_anchor":true,"ref_index":1,"title":"A General Language Assistant as a Laboratory for Alignment","work_id":"a43f9ea0-01be-47d5-b8ee-a1a9f73381c5","year":null},{"cited_arxiv_id":"1607.07086","doi":"","is_internal_anchor":true,"ref_index":2,"title":"arXiv preprint arXiv:1607.07086 , year=","work_id":"a298def3-ff4b-4b72-9b55-5707acf335ba","year":null},{"cited_arxiv_id":"2204.05862","doi":"","is_internal_anchor":true,"ref_index":3,"title":"Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback","work_id":"a1f2574b-a899-4713-be60-c87ba332656c","year":2005},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"E., and Wang, W","work_id":"63b29782-c84e-4467-9583-aee36944de57","year":null},{"cited_arxiv_id":"2208.01618","doi":"","is_internal_anchor":true,"ref_index":5,"title":"An Image is Worth One Word: Personalizing Text-to-Image Generation using Textual Inversion","work_id":"ca618c21-3ba6-448e-bd86-bcecff3cdeb5","year":null}],"snapshot_sha256":"86521c9baf0ae9ec1490431a238d8f1e0c731923e8d77dff6e41c6f4033768ab"},"source":{"id":"2302.12192","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-14T20:35:53.230683Z","id":"88c2a63e-42c0-43e5-86ce-ba86a719a11a","model_set":{"reader":"grok-4.3"},"one_line_summary":"A three-stage fine-tuning process uses human ratings to train a reward model and then improves text-to-image alignment by maximizing reward-weighted likelihood.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Fine-tuning text-to-image models with human feedback improves accuracy on prompts specifying colors, counts, and backgrounds.","strongest_claim":"Our method generates objects with specified colors, counts and backgrounds more accurately than the pre-trained model.","weakest_assumption":"Human feedback on image-text alignment is consistent enough to be captured by a learned reward function that generalizes to new prompts and does not introduce unintended biases during fine-tuning."}},"verdict_id":"88c2a63e-42c0-43e5-86ce-ba86a719a11a"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:3ce70df0fbffe49891bfd3d8d721359e9e91d0c0d8f363df014d93271f3327b7","target":"record","created_at":"2026-05-17T23:39:21Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"2c26c5b752bc8facff7879651ec9d50c096133e39800c4894b64e41096fcbb67","cross_cats_sorted":["cs.AI","cs.CV"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2023-02-23T17:34:53Z","title_canon_sha256":"b9f9d0a75704678b4acde8885bfbea84f3cb70eca4a6315ff07a9f7d2fb3b1f0"},"schema_version":"1.0","source":{"id":"2302.12192","kind":"arxiv","version":1}},"canonical_sha256":"8ba9c6f111c6767aed9f3aa6ade31ee6d7254c3ff5008fbdfaa60efc0bdb941f","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"8ba9c6f111c6767aed9f3aa6ade31ee6d7254c3ff5008fbdfaa60efc0bdb941f","first_computed_at":"2026-05-17T23:39:21.805215Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-17T23:39:21.805215Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"hEJlGv0W8AGsaypS5YJB1qcRGiTXbG8+8AENeAzCX4GQwsYfW6WSQPKRItl24i6O2yDFarRK/1nIvE6aocIBDw==","signature_status":"signed_v1","signed_at":"2026-05-17T23:39:21.805985Z","signed_message":"canonical_sha256_bytes"},"source_id":"2302.12192","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:3ce70df0fbffe49891bfd3d8d721359e9e91d0c0d8f363df014d93271f3327b7","sha256:2dd1ca0a3fd9b18f4132f26591c1606d4cff3ac4add15686da72a23d57dcfce0"],"state_sha256":"de5711bff540d77cefc0910b3e6dda7160330b8fd6999d0b0f1977c31069c6f3"}