{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:KKFBMZGHARJKFW7WUMB3VIV7G7","short_pith_number":"pith:KKFBMZGH","canonical_record":{"source":{"id":"2605.14004","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-13T18:11:16Z","cross_cats_sorted":[],"title_canon_sha256":"6241970baf562565c6fbe6ff8eb3e7e39a15c82ecca4b3d8e768a6c4604cb163","abstract_canon_sha256":"d00c9fbd3b8badf57911c87242f74136a27e3e703bd91fe3ddf9a5c9e34482c0"},"schema_version":"1.0"},"canonical_sha256":"528a1664c70452a2dbf6a303baa2bf37ed5782974cf42ef0cbb72f5223c5e8b0","source":{"kind":"arxiv","id":"2605.14004","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.14004","created_at":"2026-05-17T23:39:13Z"},{"alias_kind":"arxiv_version","alias_value":"2605.14004v1","created_at":"2026-05-17T23:39:13Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.14004","created_at":"2026-05-17T23:39:13Z"},{"alias_kind":"pith_short_12","alias_value":"KKFBMZGHARJK","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"KKFBMZGHARJKFW7W","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"KKFBMZGH","created_at":"2026-05-18T12:33:37Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:KKFBMZGHARJKFW7WUMB3VIV7G7","target":"record","payload":{"canonical_record":{"source":{"id":"2605.14004","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-13T18:11:16Z","cross_cats_sorted":[],"title_canon_sha256":"6241970baf562565c6fbe6ff8eb3e7e39a15c82ecca4b3d8e768a6c4604cb163","abstract_canon_sha256":"d00c9fbd3b8badf57911c87242f74136a27e3e703bd91fe3ddf9a5c9e34482c0"},"schema_version":"1.0"},"canonical_sha256":"528a1664c70452a2dbf6a303baa2bf37ed5782974cf42ef0cbb72f5223c5e8b0","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-17T23:39:13.129005Z","signature_b64":"jPWTVvAgVfaH7GqBhmejNF8qzxPcd1oqV79v+oPcfoQetG9t9+Plr8HyVPH2PSK8vxKcMiwYX7u5swW7a1S1BA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"528a1664c70452a2dbf6a303baa2bf37ed5782974cf42ef0cbb72f5223c5e8b0","last_reissued_at":"2026-05-17T23:39:13.128457Z","signature_status":"signed_v1","first_computed_at":"2026-05-17T23:39:13.128457Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2605.14004","source_version":1,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-17T23:39:13Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"A7cMjOGiuRJHCksR2NQM3iW1Ll2jTT0+DqBQYeTIQFlElq4RDE370ROuFsbx6pPmC2rnz1Nfi4dU7GOv4dLfCg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-27T07:39:45.105204Z"},"content_sha256":"2141b0684a9f4d65831b6dce4d2312d23b24eca84594819f983060f0ca6c0eeb","schema_version":"1.0","event_id":"sha256:2141b0684a9f4d65831b6dce4d2312d23b24eca84594819f983060f0ca6c0eeb"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:KKFBMZGHARJKFW7WUMB3VIV7G7","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Conditional Attribute Estimation with Autoregressive Sequence Models","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"Conditional Attribute Transformers estimate sequence attributes from each possible next token in one forward pass.","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Andrew J. Loza, Daniella Meeker, Erica Stutz, Giacomo Marino, Qiao Liu","submitted_at":"2026-05-13T18:11:16Z","abstract_excerpt":"Generative models are often trained with a next-token prediction objective, yet many downstream applications require the ability to estimate or control sequence-level properties. Next-token prediction can lead to overfitting of local patterns during training, underfitting of global structure, and requires significant downstream modifications or expensive sampling to guide or predict the global attributes of generated samples at inference time. Here, we introduce Conditional Attribute Transformers, a novel method for jointly estimating the next-token probability and the value of an attribute co"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Our approach achieves state of the art performance on sparse reward tasks, improves next-token prediction at sufficient model sizes, estimates attribute probabilities orders of magnitude faster than sampling, and can guide decoding of autoregressive sequence models on a range of language tasks.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The framework assumes that sequence-level attributes can be accurately estimated from partial sequences and single next-token conditionals without requiring full-sequence rollouts or additional supervision during training.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Conditional Attribute Transformers jointly estimate next-token probabilities and conditional attribute values for autoregressive sequence models, enabling credit assignment, counterfactuals, and steerable generation in one pass.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Conditional Attribute Transformers estimate sequence attributes from each possible next token in one forward pass.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"53a355f7c4715abfb75857a94a2b71bebb28e87d50e19073f0896d96cdf975a0"},"source":{"id":"2605.14004","kind":"arxiv","version":1},"verdict":{"id":"66c8d1f5-1b08-4495-a554-38288d64531b","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T05:46:08.145797Z","strongest_claim":"Our approach achieves state of the art performance on sparse reward tasks, improves next-token prediction at sufficient model sizes, estimates attribute probabilities orders of magnitude faster than sampling, and can guide decoding of autoregressive sequence models on a range of language tasks.","one_line_summary":"Conditional Attribute Transformers jointly estimate next-token probabilities and conditional attribute values for autoregressive sequence models, enabling credit assignment, counterfactuals, and steerable generation in one pass.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The framework assumes that sequence-level attributes can be accurately estimated from partial sequences and single next-token conditionals without requiring full-sequence rollouts or additional supervision during training.","pith_extraction_headline":"Conditional Attribute Transformers estimate sequence attributes from each possible next token in one forward pass."},"references":{"count":37,"sample":[{"doi":"","year":1901,"title":"Language models are few-shot learners.Advances in neural information processing systems, 33:1877–1901","work_id":"bd7ec542-9242-446e-955e-bb75e729be5d","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2022,"title":"Protgpt2 is a deep unsupervised language model for protein design.Nature communications, 13(1):4348","work_id":"52989572-5dbc-4208-817f-d97ac7837aef","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"Generative medical event models improve with scale","work_id":"13b807b6-b6a4-434d-87e6-4342317e3090","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"Genome modeling and design across all domains of life with evo 2.BioRxiv, pages 2025–02, 2025","work_id":"0593bc66-e1d3-4563-878b-6132617f61b4","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2022,"title":"Training Compute-Optimal Large Language Models","work_id":"b2faf28d-86b7-429c-bc42-469458efc246","ref_index":5,"cited_arxiv_id":"2203.15556","is_internal_anchor":true}],"resolved_work":37,"snapshot_sha256":"dacb339e391a59586e6c054b4f54bd114cd384b91ed8cf4d3c0ad3a41d137595","internal_anchors":4},"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"},"verdict_id":"66c8d1f5-1b08-4495-a554-38288d64531b"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-17T23:39:13Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"xbCZBDCai5VlkFgbajjCfjw27EXVyR6tK7Se1DEw2AfCeYv/NetCfCwsOGNKLHQp0k8S5SAWy7ygzU02ZYYPBw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-27T07:39:45.105913Z"},"content_sha256":"06a2dee7f4dbaa313d60a68a6d035020794ea29ee78fcf309028655d20c35f7d","schema_version":"1.0","event_id":"sha256:06a2dee7f4dbaa313d60a68a6d035020794ea29ee78fcf309028655d20c35f7d"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/KKFBMZGHARJKFW7WUMB3VIV7G7/bundle.json","state_url":"https://pith.science/pith/KKFBMZGHARJKFW7WUMB3VIV7G7/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/KKFBMZGHARJKFW7WUMB3VIV7G7/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-05-27T07:39:45Z","links":{"resolver":"https://pith.science/pith/KKFBMZGHARJKFW7WUMB3VIV7G7","bundle":"https://pith.science/pith/KKFBMZGHARJKFW7WUMB3VIV7G7/bundle.json","state":"https://pith.science/pith/KKFBMZGHARJKFW7WUMB3VIV7G7/state.json","well_known_bundle":"https://pith.science/.well-known/pith/KKFBMZGHARJKFW7WUMB3VIV7G7/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:KKFBMZGHARJKFW7WUMB3VIV7G7","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":"d00c9fbd3b8badf57911c87242f74136a27e3e703bd91fe3ddf9a5c9e34482c0","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-13T18:11:16Z","title_canon_sha256":"6241970baf562565c6fbe6ff8eb3e7e39a15c82ecca4b3d8e768a6c4604cb163"},"schema_version":"1.0","source":{"id":"2605.14004","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.14004","created_at":"2026-05-17T23:39:13Z"},{"alias_kind":"arxiv_version","alias_value":"2605.14004v1","created_at":"2026-05-17T23:39:13Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.14004","created_at":"2026-05-17T23:39:13Z"},{"alias_kind":"pith_short_12","alias_value":"KKFBMZGHARJK","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"KKFBMZGHARJKFW7W","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"KKFBMZGH","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:06a2dee7f4dbaa313d60a68a6d035020794ea29ee78fcf309028655d20c35f7d","target":"graph","created_at":"2026-05-17T23:39:13Z","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 approach achieves state of the art performance on sparse reward tasks, improves next-token prediction at sufficient model sizes, estimates attribute probabilities orders of magnitude faster than sampling, and can guide decoding of autoregressive sequence models on a range of language tasks."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"The framework assumes that sequence-level attributes can be accurately estimated from partial sequences and single next-token conditionals without requiring full-sequence rollouts or additional supervision during training."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"Conditional Attribute Transformers jointly estimate next-token probabilities and conditional attribute values for autoregressive sequence models, enabling credit assignment, counterfactuals, and steerable generation in one pass."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Conditional Attribute Transformers estimate sequence attributes from each possible next token in one forward pass."}],"snapshot_sha256":"53a355f7c4715abfb75857a94a2b71bebb28e87d50e19073f0896d96cdf975a0"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"paper":{"abstract_excerpt":"Generative models are often trained with a next-token prediction objective, yet many downstream applications require the ability to estimate or control sequence-level properties. Next-token prediction can lead to overfitting of local patterns during training, underfitting of global structure, and requires significant downstream modifications or expensive sampling to guide or predict the global attributes of generated samples at inference time. Here, we introduce Conditional Attribute Transformers, a novel method for jointly estimating the next-token probability and the value of an attribute co","authors_text":"Andrew J. Loza, Daniella Meeker, Erica Stutz, Giacomo Marino, Qiao Liu","cross_cats":[],"headline":"Conditional Attribute Transformers estimate sequence attributes from each possible next token in one forward pass.","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-13T18:11:16Z","title":"Conditional Attribute Estimation with Autoregressive Sequence Models"},"references":{"count":37,"internal_anchors":4,"resolved_work":37,"sample":[{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":1,"title":"Language models are few-shot learners.Advances in neural information processing systems, 33:1877–1901","work_id":"bd7ec542-9242-446e-955e-bb75e729be5d","year":1901},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"Protgpt2 is a deep unsupervised language model for protein design.Nature communications, 13(1):4348","work_id":"52989572-5dbc-4208-817f-d97ac7837aef","year":2022},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"Generative medical event models improve with scale","work_id":"13b807b6-b6a4-434d-87e6-4342317e3090","year":2025},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"Genome modeling and design across all domains of life with evo 2.BioRxiv, pages 2025–02, 2025","work_id":"0593bc66-e1d3-4563-878b-6132617f61b4","year":2025},{"cited_arxiv_id":"2203.15556","doi":"","is_internal_anchor":true,"ref_index":5,"title":"Training Compute-Optimal Large Language Models","work_id":"b2faf28d-86b7-429c-bc42-469458efc246","year":2022}],"snapshot_sha256":"dacb339e391a59586e6c054b4f54bd114cd384b91ed8cf4d3c0ad3a41d137595"},"source":{"id":"2605.14004","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-15T05:46:08.145797Z","id":"66c8d1f5-1b08-4495-a554-38288d64531b","model_set":{"reader":"grok-4.3"},"one_line_summary":"Conditional Attribute Transformers jointly estimate next-token probabilities and conditional attribute values for autoregressive sequence models, enabling credit assignment, counterfactuals, and steerable generation in one pass.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Conditional Attribute Transformers estimate sequence attributes from each possible next token in one forward pass.","strongest_claim":"Our approach achieves state of the art performance on sparse reward tasks, improves next-token prediction at sufficient model sizes, estimates attribute probabilities orders of magnitude faster than sampling, and can guide decoding of autoregressive sequence models on a range of language tasks.","weakest_assumption":"The framework assumes that sequence-level attributes can be accurately estimated from partial sequences and single next-token conditionals without requiring full-sequence rollouts or additional supervision during training."}},"verdict_id":"66c8d1f5-1b08-4495-a554-38288d64531b"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:2141b0684a9f4d65831b6dce4d2312d23b24eca84594819f983060f0ca6c0eeb","target":"record","created_at":"2026-05-17T23:39:13Z","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":"d00c9fbd3b8badf57911c87242f74136a27e3e703bd91fe3ddf9a5c9e34482c0","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-13T18:11:16Z","title_canon_sha256":"6241970baf562565c6fbe6ff8eb3e7e39a15c82ecca4b3d8e768a6c4604cb163"},"schema_version":"1.0","source":{"id":"2605.14004","kind":"arxiv","version":1}},"canonical_sha256":"528a1664c70452a2dbf6a303baa2bf37ed5782974cf42ef0cbb72f5223c5e8b0","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"528a1664c70452a2dbf6a303baa2bf37ed5782974cf42ef0cbb72f5223c5e8b0","first_computed_at":"2026-05-17T23:39:13.128457Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-17T23:39:13.128457Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"jPWTVvAgVfaH7GqBhmejNF8qzxPcd1oqV79v+oPcfoQetG9t9+Plr8HyVPH2PSK8vxKcMiwYX7u5swW7a1S1BA==","signature_status":"signed_v1","signed_at":"2026-05-17T23:39:13.129005Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.14004","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:2141b0684a9f4d65831b6dce4d2312d23b24eca84594819f983060f0ca6c0eeb","sha256:06a2dee7f4dbaa313d60a68a6d035020794ea29ee78fcf309028655d20c35f7d"],"state_sha256":"70649a2853fd637936fb467ffe788a4068d0f2d5a5ebf4dd2800ca291a35dc33"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"sug2Nlx9jRJELY6s/uiHe+9uT9716NCz1GcXC/bgvRNiT3Mw5ohOl1BczTXdZME9L3iWoA6iLZYoKI5evdRyCg==","signed_message":"bundle_sha256_bytes","signed_at":"2026-05-27T07:39:45.109701Z","bundle_sha256":"2aee5347ff059c554f413b138c4c539ddd045df63f57683ac0c83e370da7ecf6"}}