{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:IATM46SH4DVYTKK5TNAGKNUSLU","short_pith_number":"pith:IATM46SH","schema_version":"1.0","canonical_sha256":"4026ce7a47e0eb89a95d9b406536925d22b467f63c263f343b9c783534b813f3","source":{"kind":"arxiv","id":"2201.08239","version":3},"attestation_state":"computed","paper":{"title":"LaMDA: Language Models for Dialog Applications","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Fine-tuning LaMDA models on annotated human values plus access to external tools markedly raises safety and factual grounding in dialog responses.","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Aaron Cohen, Adam Roberts, Alejandra Molina, Alena Butryna, Alicia Jin, Amin Ghafouri, Apoorv Kulshreshtha, Ben Hutchinson, Ben Zevenbergen, Blaise Aguera-Arcas, Chung-Ching Chang, Claire Cui, Daniel De Freitas, Dehao Chen, Dmitry Lepikhin, Ed Chi, Erin Hoffman-John, Heng-Tze Cheng, Hongrae Lee, Huaixiu Steven Zheng, Igor Krivokon, James Qin, Jamie Hall, Joe Fenton, Johnny Soraker, Josh Lee, Kathleen Meier-Hellstern, Kristen Olson, Laichee Man, Leslie Baker, Lora Aroyo, Maarten Bosma, Marcelo Menegali, Marc Pickett, Marian Croak, Mark Diaz, Matthew Lamm, Maxim Krikun, Meredith Ringel Morris, Noam Shazeer, Pranesh Srinivasan, Quoc Le, Rachel Bernstein, Ravi Rajakumar, Ray Kurzweil, Renelito Delos Santos, Romal Thoppilan, Taylor Bos, Toju Duke, Tulsee Doshi, Viktoriya Kuzmina, Vincent Zhao, Vinodkumar Prabhakaran, Will Rusch, Yaguang Li, Yanping Huang, Yanqi Zhou, Yuanzhong Xu, Yu Du, Zhifeng Chen","submitted_at":"2022-01-20T15:44:37Z","abstract_excerpt":"We present LaMDA: Language Models for Dialog Applications. LaMDA is a family of Transformer-based neural language models specialized for dialog, which have up to 137B parameters and are pre-trained on 1.56T words of public dialog data and web text. While model scaling alone can improve quality, it shows less improvements on safety and factual grounding. We demonstrate that fine-tuning with annotated data and enabling the model to consult external knowledge sources can lead to significant improvements towards the two key challenges of safety and factual grounding. The first challenge, safety, i"},"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":true,"formal_links_present":false},"canonical_record":{"source":{"id":"2201.08239","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2022-01-20T15:44:37Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"7eb86eef1745306f09fec4112d4b5fe83e79a6ddecbb9cbe3dee886ae97357dd","abstract_canon_sha256":"892b8542925d98985ea67a8ae2d0bd8a9db63b98cc9737b2a1e4216b953ea86b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T03:55:51.905581Z","signature_b64":"q1gvmEYPfcrJHV6hZ4UDBrgz9Njn/VhRQD1q4cncEYiDQxNtXpNFerJEWEDE1XfwWmRJzT/AgsdwwsUp4f2XAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"4026ce7a47e0eb89a95d9b406536925d22b467f63c263f343b9c783534b813f3","last_reissued_at":"2026-07-05T03:55:51.905047Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T03:55:51.905047Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"LaMDA: Language Models for Dialog Applications","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Fine-tuning LaMDA models on annotated human values plus access to external tools markedly raises safety and factual grounding in dialog responses.","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Aaron Cohen, Adam Roberts, Alejandra Molina, Alena Butryna, Alicia Jin, Amin Ghafouri, Apoorv Kulshreshtha, Ben Hutchinson, Ben Zevenbergen, Blaise Aguera-Arcas, Chung-Ching Chang, Claire Cui, Daniel De Freitas, Dehao Chen, Dmitry Lepikhin, Ed Chi, Erin Hoffman-John, Heng-Tze Cheng, Hongrae Lee, Huaixiu Steven Zheng, Igor Krivokon, James Qin, Jamie Hall, Joe Fenton, Johnny Soraker, Josh Lee, Kathleen Meier-Hellstern, Kristen Olson, Laichee Man, Leslie Baker, Lora Aroyo, Maarten Bosma, Marcelo Menegali, Marc Pickett, Marian Croak, Mark Diaz, Matthew Lamm, Maxim Krikun, Meredith Ringel Morris, Noam Shazeer, Pranesh Srinivasan, Quoc Le, Rachel Bernstein, Ravi Rajakumar, Ray Kurzweil, Renelito Delos Santos, Romal Thoppilan, Taylor Bos, Toju Duke, Tulsee Doshi, Viktoriya Kuzmina, Vincent Zhao, Vinodkumar Prabhakaran, Will Rusch, Yaguang Li, Yanping Huang, Yanqi Zhou, Yuanzhong Xu, Yu Du, Zhifeng Chen","submitted_at":"2022-01-20T15:44:37Z","abstract_excerpt":"We present LaMDA: Language Models for Dialog Applications. LaMDA is a family of Transformer-based neural language models specialized for dialog, which have up to 137B parameters and are pre-trained on 1.56T words of public dialog data and web text. While model scaling alone can improve quality, it shows less improvements on safety and factual grounding. We demonstrate that fine-tuning with annotated data and enabling the model to consult external knowledge sources can lead to significant improvements towards the two key challenges of safety and factual grounding. The first challenge, safety, i"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"fine-tuning with annotated data and enabling the model to consult external knowledge sources can lead to significant improvements towards the two key challenges of safety and factual grounding.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The assumption that the illustrative set of human values used for annotation and the chosen external knowledge sources (IR, translator, calculator) are sufficient to capture the full range of safety and factuality requirements in open-ended real-world dialogs.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"LaMDA shows that fine-tuning on human-value annotations and consulting external knowledge sources significantly improves safety and factual grounding in large dialog models beyond what scaling alone achieves.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Fine-tuning LaMDA models on annotated human values plus access to external tools markedly raises safety and factual grounding in dialog responses.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"6073724fbf62453f00844560f7b8b585cfb8e9904408f340905dacb901214118"},"source":{"id":"2201.08239","kind":"arxiv","version":3},"verdict":{"id":"cff0969a-8dd4-474a-a31c-54a266139c3b","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-12T03:09:49.108188Z","strongest_claim":"fine-tuning with annotated data and enabling the model to consult external knowledge sources can lead to significant improvements towards the two key challenges of safety and factual grounding.","one_line_summary":"LaMDA shows that fine-tuning on human-value annotations and consulting external knowledge sources significantly improves safety and factual grounding in large dialog models beyond what scaling alone achieves.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The assumption that the illustrative set of human values used for annotation and the chosen external knowledge sources (IR, translator, calculator) are sufficient to capture the full range of safety and factuality requirements in open-ended real-world dialogs.","pith_extraction_headline":"Fine-tuning LaMDA models on annotated human values plus access to external tools markedly raises safety and factual grounding in dialog responses."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2201.08239/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":120,"sample":[{"doi":"","year":2015,"title":"Skip-thought vectors","work_id":"9cece506-80e6-45d7-8b11-34a3b1eb1188","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2015,"title":"Semi-supervised sequence learning","work_id":"73fcad66-0e78-4869-8221-6372d91a4b6c","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2018,"title":"Deep contextualized word representations","work_id":"274f7727-8098-46a2-8cdc-e644ce291052","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2018,"title":"Improving language understanding by generative pre-training","work_id":"ffbc73cf-9e41-4b1b-bdaf-3fa451b93493","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2019,"title":"BERT: Pre-training of deep bidirectional transformers for language understanding","work_id":"df8eae13-96c1-4b4c-8d31-f4b903607b72","ref_index":6,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":120,"snapshot_sha256":"06393da193bdeead745e170832bfadaa81ff15b16e7d05e99d782f2782ebffa8","internal_anchors":15},"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":"2201.08239","created_at":"2026-07-05T03:55:51.905120+00:00"},{"alias_kind":"arxiv_version","alias_value":"2201.08239v3","created_at":"2026-07-05T03:55:51.905120+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2201.08239","created_at":"2026-07-05T03:55:51.905120+00:00"},{"alias_kind":"pith_short_12","alias_value":"IATM46SH4DVY","created_at":"2026-07-05T03:55:51.905120+00:00"},{"alias_kind":"pith_short_16","alias_value":"IATM46SH4DVYTKK5","created_at":"2026-07-05T03:55:51.905120+00:00"},{"alias_kind":"pith_short_8","alias_value":"IATM46SH","created_at":"2026-07-05T03:55:51.905120+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":88,"internal_anchor_count":63,"sample":[{"citing_arxiv_id":"2606.26981","citing_title":"In-Context Model Predictive Generation: Open-Vocabulary Motion Synthesis from Language Models to Physics","ref_index":13,"is_internal_anchor":true},{"citing_arxiv_id":"2606.26587","citing_title":"SharQ: Bridging Activation Sparsity and FP4 Quantization for LLM Inference","ref_index":120,"is_internal_anchor":true},{"citing_arxiv_id":"2606.11886","citing_title":"Real-Time Language Model Jamming: A Case Study for Live Music Accompaniment Generation","ref_index":5,"is_internal_anchor":true},{"citing_arxiv_id":"2606.11953","citing_title":"Decoding Multimodal Cues: Unveiling the Implicit Meaning Behind Hateful Videos","ref_index":32,"is_internal_anchor":true},{"citing_arxiv_id":"2606.08410","citing_title":"Provably Efficient Personalized Multi-Objective Bandits with Proactive Conversational Queries","ref_index":80,"is_internal_anchor":true},{"citing_arxiv_id":"2606.04661","citing_title":"CRAFT: Cost-aware Refinement And Front-aware Tuning of Prompts","ref_index":163,"is_internal_anchor":true},{"citing_arxiv_id":"2606.04945","citing_title":"STaR-Quant: State-Time Consistent Post-Training Quantization for Diffusion Large Language Models","ref_index":123,"is_internal_anchor":true},{"citing_arxiv_id":"2606.30642","citing_title":"LeVo 2: Stable and Melodious Song Generation via Hierarchical Representation Modeling and Progressive Post-Training","ref_index":34,"is_internal_anchor":true},{"citing_arxiv_id":"2606.30783","citing_title":"Security--Fidelity Tradeoffs: The Hidden Cost of Prompt Injection Defense","ref_index":104,"is_internal_anchor":true},{"citing_arxiv_id":"2605.27963","citing_title":"Throughput-Optimized Networks at Scale","ref_index":87,"is_internal_anchor":true},{"citing_arxiv_id":"2606.27981","citing_title":"ToxiREX: A Dataset on Toxic REasoning in ConteXt","ref_index":174,"is_internal_anchor":true},{"citing_arxiv_id":"2302.12039","citing_title":"Natural Language Processing in the Legal Domain","ref_index":36,"is_internal_anchor":true},{"citing_arxiv_id":"2301.13688","citing_title":"The Flan Collection: Designing Data and Methods for Effective Instruction Tuning","ref_index":58,"is_internal_anchor":true},{"citing_arxiv_id":"2309.16609","citing_title":"Qwen Technical Report","ref_index":5,"is_internal_anchor":true},{"citing_arxiv_id":"2312.11805","citing_title":"Gemini: A Family of Highly Capable Multimodal Models","ref_index":104,"is_internal_anchor":true},{"citing_arxiv_id":"2305.09617","citing_title":"Towards Expert-Level Medical Question Answering with Large Language Models","ref_index":2,"is_internal_anchor":true},{"citing_arxiv_id":"2409.10102","citing_title":"Trustworthiness in Retrieval-Augmented Generation Systems: A Survey","ref_index":133,"is_internal_anchor":true},{"citing_arxiv_id":"2305.02301","citing_title":"Distilling Step-by-Step! Outperforming Larger Language Models with Less Training Data and Smaller Model Sizes","ref_index":102,"is_internal_anchor":true},{"citing_arxiv_id":"2602.03433","citing_title":"When control meets large language models: From words to dynamics","ref_index":16,"is_internal_anchor":true},{"citing_arxiv_id":"2605.18763","citing_title":"Query-Conditioned Graph Retrieval for Contextualized LLM Reasoning in Personalized Wearable Data","ref_index":6,"is_internal_anchor":true},{"citing_arxiv_id":"2307.06435","citing_title":"A Comprehensive Overview of Large Language Models","ref_index":150,"is_internal_anchor":true},{"citing_arxiv_id":"2506.00166","citing_title":"Disentangled Safety Adapters Enable Efficient Guardrails and Flexible Inference-Time Alignment","ref_index":43,"is_internal_anchor":true},{"citing_arxiv_id":"2310.10631","citing_title":"Llemma: An Open Language Model For Mathematics","ref_index":184,"is_internal_anchor":true},{"citing_arxiv_id":"2408.00724","citing_title":"Inference Scaling Laws: An Empirical Analysis of Compute-Optimal Inference for Problem-Solving with Language Models","ref_index":254,"is_internal_anchor":true},{"citing_arxiv_id":"2305.16264","citing_title":"Scaling Data-Constrained Language Models","ref_index":117,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/IATM46SH4DVYTKK5TNAGKNUSLU","json":"https://pith.science/pith/IATM46SH4DVYTKK5TNAGKNUSLU.json","graph_json":"https://pith.science/api/pith-number/IATM46SH4DVYTKK5TNAGKNUSLU/graph.json","events_json":"https://pith.science/api/pith-number/IATM46SH4DVYTKK5TNAGKNUSLU/events.json","paper":"https://pith.science/paper/IATM46SH"},"agent_actions":{"view_html":"https://pith.science/pith/IATM46SH4DVYTKK5TNAGKNUSLU","download_json":"https://pith.science/pith/IATM46SH4DVYTKK5TNAGKNUSLU.json","view_paper":"https://pith.science/paper/IATM46SH","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2201.08239&json=true","fetch_graph":"https://pith.science/api/pith-number/IATM46SH4DVYTKK5TNAGKNUSLU/graph.json","fetch_events":"https://pith.science/api/pith-number/IATM46SH4DVYTKK5TNAGKNUSLU/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/IATM46SH4DVYTKK5TNAGKNUSLU/action/timestamp_anchor","attest_storage":"https://pith.science/pith/IATM46SH4DVYTKK5TNAGKNUSLU/action/storage_attestation","attest_author":"https://pith.science/pith/IATM46SH4DVYTKK5TNAGKNUSLU/action/author_attestation","sign_citation":"https://pith.science/pith/IATM46SH4DVYTKK5TNAGKNUSLU/action/citation_signature","submit_replication":"https://pith.science/pith/IATM46SH4DVYTKK5TNAGKNUSLU/action/replication_record"}},"created_at":"2026-07-05T03:55:51.905120+00:00","updated_at":"2026-07-05T03:55:51.905120+00:00"}