{"work":{"id":"62105b61-411f-46e5-970a-bd322c6adbbc","openalex_id":"https://openalex.org/W4378771755","doi":"10.48550/arxiv.2305.18290","arxiv_id":"2305.18290","raw_key":null,"title":"Direct Preference Optimization: Your Language Model is Secretly a Reward Model","authors":null,"authors_text":"Rafael Rafailov, Archit Sharma, Eric Mitchell, Stefano Ermon, Christopher D. Manning, Chelsea Finn","year":2023,"venue":"cs.LG","abstract":"While large-scale unsupervised language models (LMs) learn broad world knowledge and some reasoning skills, achieving precise control of their behavior is difficult due to the completely unsupervised nature of their training. Existing methods for gaining such steerability collect human labels of the relative quality of model generations and fine-tune the unsupervised LM to align with these preferences, often with reinforcement learning from human feedback (RLHF). However, RLHF is a complex and often unstable procedure, first fitting a reward model that reflects the human preferences, and then fine-tuning the large unsupervised LM using reinforcement learning to maximize this estimated reward without drifting too far from the original model. In this paper we introduce a new parameterization of the reward model in RLHF that enables extraction of the corresponding optimal policy in closed form, allowing us to solve the standard RLHF problem with only a simple classification loss. The resulting algorithm, which we call Direct Preference Optimization (DPO), is stable, performant, and computationally lightweight, eliminating the need for sampling from the LM during fine-tuning or performing significant hyperparameter tuning. Our experiments show that DPO can fine-tune LMs to align with human preferences as well as or better than existing methods. Notably, fine-tuning with DPO exceeds PPO-based RLHF in ability to control sentiment of generations, and matches or improves response quality in summarization and single-turn dialogue while being substantially simpler to implement and train.","external_url":"https://arxiv.org/abs/2305.18290","cited_by_count":281,"metadata_source":"pith","metadata_fetched_at":"2026-08-05T02:28:24.338817+00:00","pith_arxiv_id":"2305.18290","created_at":"2026-05-09T06:20:41.745842+00:00","updated_at":"2026-08-05T02:28:24.338817+00:00","title_quality_ok":true,"display_title":"Direct Preference Optimization: Your Language Model is Secretly a Reward Model","render_title":"Direct Preference Optimization: Your Language Model is Secretly a Reward Model"},"hub":{"state":{"work_id":"62105b61-411f-46e5-970a-bd322c6adbbc","tier":"super_hub","tier_reason":"100+ Pith inbound or 10,000+ external citations","pith_inbound_count":181,"external_cited_by_count":281,"distinct_field_count":19,"first_pith_cited_at":"2023-07-12T20:01:52+00:00","last_pith_cited_at":"2026-07-09T00:27:07+00:00","author_build_status":"needed","summary_status":"needed","contexts_status":"needed","graph_status":"needed","ask_index_status":"needed","reader_status":"not_needed","recognition_status":"not_needed","updated_at":"2026-08-23T17:09:23.293272+00:00","tier_text":"super_hub"},"tier":"super_hub","role_counts":[{"context_role":"background","n":26},{"context_role":"method","n":9},{"context_role":"baseline","n":1},{"context_role":"dataset","n":1}],"polarity_counts":[{"context_polarity":"background","n":25},{"context_polarity":"use_method","n":9},{"context_polarity":"baseline","n":1},{"context_polarity":"unclear","n":1},{"context_polarity":"use_dataset","n":1}],"runs":{"ask_index":{"job_type":"ask_index","status":"succeeded","result":{"title":"Direct Preference Optimization: Your Language Model is Secretly a Reward Model","claims":[{"claim_text":"While large-scale unsupervised language models (LMs) learn broad world knowledge and some reasoning skills, achieving precise control of their behavior is difficult due to the completely unsupervised nature of their training. Existing methods for gaining such steerability collect human labels of the relative quality of model generations and fine-tune the unsupervised LM to align with these preferences, often with reinforcement learning from human feedback (RLHF). However, RLHF is a complex and often unstable procedure, first fitting a reward model that reflects the human preferences, and then ","claim_type":"abstract","evidence_strength":"source_metadata"},{"claim_text":"Logs Backward Slicing o3-mini, DeepSeek LLaMA3-70B PA, L-ACC AUCAD [198] Align Gen Fine-tuningMethod + Is- sue Logs Issue Alignment LLaMA3.1 Magicoder PA, LA, VF1 Zhong et al. [214] Full Gen Fine-tuning Java Method Logs Compact Model FTLLaMA-8B, Mistral GPT-4o, Claude PA, LA, F1 Specialized Logging Generation (Test Code, File-level, Configuration) File-Level [140] File Gen ICL Python File Multi Logs Pipeline Stages GPT-4o-mini - L-ACC, AOD Test-Code [151] Test Log ICL Test Method Assert/Log Test","claim_type":"background","confidence":0.9,"evidence_strength":"citation_context"},{"claim_text":"resulting in reward hacking. We address (1) by grounding the entire framework on diverse text sampled from a content-rich pre-training corpus. We ask the model to synthesize queries that draw from the text, and, in tasks for which ground truth exists, to construct queries whose answers are derivable from the text. We address (2) by using Direct Preference Optimization (DPO) [ 28], to train the LLM via contrastive signals from the response pairs. We use two methods to mitigate (3). First, we give","claim_type":"method","confidence":0.9,"evidence_strength":"citation_context"},{"claim_text":"pLogP, and QED), the LLM-based models consistently achieve much higher similarity to the source molecule, with SIM typically in the range of 0.48-0.59, whereas the non-LLM baselines remain in a much lower range 11 Table 6: Comparison with representative non-LLM baselines under a similarity-aware PMO objective. The main comparison uses a generation budget of 20. Method DRD2 GSK3β pLogP QED SR↑SIM↑RI↑ SR↑SIM↑RI↑ SR↑SIM↑RI↑ SR↑SIM↑RI↑ GraphGA (20) 30.80 0.210 12.86 100.000.077 15.48 100.000.154 11.","claim_type":"baseline","confidence":0.9,"evidence_strength":"citation_context"},{"claim_text":"Despite their impressive fluency, LLMs struggle with hallucination, generat- ing text that is plausible but factually ungrounded [17,19]. When extended to the visual domain, VLMs inherit this vulnerability and exhibit visual halluci- nation-confidently describing entities, attributes, or actions that are entirely absent from the visual input even if semantically plausible [15,29,38]. Conse- quently, as models evolve to process increasingly diverse sensory streams, main- taining strict factual gr","claim_type":"background","confidence":0.9,"evidence_strength":"citation_context"},{"claim_text":"where πθ(y|x, ϕ) denotes the policy induced by parametersθ when conditioned on textual contextϕ and queryx. For a task distributionDand rewardr, the natural joint objective is max θ, ϕ J(θ, ϕ) =E x∼D, y∼π θ(·|x,ϕ) \u0002 r(x, y) \u0003 .(2) Each factor admits many concrete optimizers. On the slow side,θ can be updated by SFT, preference optimization [43], or policy-gradient methods such as PPO [46] and GRPO [47], frequently under verifiable rewards [26]. On the fast side,ϕ can be updated by automated prom","claim_type":"method","confidence":0.9,"evidence_strength":"citation_context"},{"claim_text":"Commonsense Reasoning: We included PIQA[4], SIQA[63], HellaSwag[89], WinoGrande [62], ARC[11], OpenBookQA(OBQA)[50], and CommonsenseQA(CSQA)[ 75] to assess common sense reasoning. CSQA was exclusively tested using a 7-shot setup, while all other tests were conducted with a 0-shot configuration. Reading Comprehension: For reading comprehension, we report the 0-shot average on SQuAD[61], QuAC[8], and BoolQ[10]. Math: We report the average of the GSM8K[12] (8 shot), and MATH[28] (4 shot) benchmarks","claim_type":"dataset","confidence":0.9,"evidence_strength":"citation_context"}],"why_cited":"Pith tracks Direct Preference Optimization: Your Language Model is Secretly a Reward Model because it crossed a citation-hub threshold. Current citing contexts most often use it as background evidence (20 contexts).","role_counts":[{"n":20,"context_role":"background"},{"n":8,"context_role":"method"},{"n":1,"context_role":"baseline"},{"n":1,"context_role":"dataset"}]},"error":null,"updated_at":"2026-05-21T05:42:37.639284+00:00"},"author_expand":{"job_type":"author_expand","status":"succeeded","result":{"authors_linked":[{"id":"bafd1009-1b9f-4624-8ceb-ecf73ce7f982","orcid":null,"display_name":"Rafael Rafailov"},{"id":"52423bdb-17c9-48b5-a88d-611abbecf1fb","orcid":null,"display_name":"Archit Sharma"},{"id":"bebec598-67af-4cdc-8623-d141039043ed","orcid":null,"display_name":"Eric Mitchell"},{"id":"e00f90b7-aa2a-4902-8d30-19cdf0a496f9","orcid":null,"display_name":"Stefano Ermon"},{"id":"d649e132-c634-4aa5-9f58-83571eb7fc75","orcid":null,"display_name":"Christopher D. Manning"},{"id":"379e406e-0cbc-4ede-b9dd-9a76a16a6da8","orcid":null,"display_name":"Chelsea Finn"}]},"error":null,"updated_at":"2026-05-21T05:42:38.320143+00:00"},"context_extract":{"job_type":"context_extract","status":"succeeded","result":{"enqueued_papers":25},"error":null,"updated_at":"2026-05-14T10:49:25.373653+00:00"},"graph_features":{"job_type":"graph_features","status":"succeeded","result":{"co_cited":[{"title":"Proximal Policy Optimization Algorithms","work_id":"240c67fe-d14d-4520-91c1-38a4e272ca19","shared_citers":16},{"title":"DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models","work_id":"c5006563-f3ec-438a-9e35-b7b484f34828","shared_citers":15},{"title":"Training language models to follow instructions with human feedback","work_id":"52aff42f-4fa9-4fcf-bdb3-1459b9bebf65","shared_citers":12},{"title":"Training Verifiers to Solve Math Word Problems","work_id":"acab1aa8-b4d6-40e0-a3ee-25341701dca2","shared_citers":12},{"title":"Qwen3 Technical Report","work_id":"25a4e30c-1232-48e7-9925-02fa12ba7c9e","shared_citers":11},{"title":"GPT-4 Technical Report","work_id":"b928e041-6991-4c08-8c81-0359e4097c7b","shared_citers":10},{"title":"The Llama 3 Herd of Models","work_id":"1549a635-88af-4ac1-acfe-51ae7bb53345","shared_citers":10},{"title":"Evaluating Large Language Models Trained on Code","work_id":"042493e9-b26f-4b4e-bbde-382072ca9b08","shared_citers":9},{"title":"Llama 2: Open Foundation and Fine-Tuned Chat Models","work_id":"68a5177f-d644-44c1-bd4f-4e5278c22f5d","shared_citers":8},{"title":"LLaMA: Open and Efficient Foundation Language Models","work_id":"c018fc23-6f3f-4035-9d02-28a2173b2b9d","shared_citers":8},{"title":"Constitutional AI: Harmlessness from AI Feedback","work_id":"faaaa4e0-2676-4fac-a0b4-99aef10d2095","shared_citers":7},{"title":"DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning","work_id":"e6b75ad5-2877-4168-97c8-710407094d20","shared_citers":7},{"title":"LoRA: Low-Rank Adaptation of Large Language Models","work_id":"0426219a-789e-4964-adc8-a04538510818","shared_citers":7},{"title":"Measuring Massive Multitask Language Understanding","work_id":"e87ec49a-544b-4ec8-8991-75298c64ff5e","shared_citers":7},{"title":"DAPO: An Open-Source LLM Reinforcement Learning System at Scale","work_id":"64019d00-0b11-4bbd-b173-b46c8fad0157","shared_citers":6},{"title":"Gemini: A Family of Highly Capable Multimodal Models","work_id":"83f7c85b-3f11-450f-ac0c-64d9745220b2","shared_citers":6},{"title":"Language Models are Few-Shot Learners","work_id":"214732c0-2edd-44a0-af9e-28184a2b8279","shared_citers":6},{"title":"Program Synthesis with Large Language Models","work_id":"fd241a05-03b9-4de2-9588-9d77ce176125","shared_citers":6},{"title":"Scaling Instruction-Finetuned Language Models","work_id":"8405abb1-7558-4fdf-af24-f4c52fa77a06","shared_citers":6},{"title":"Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback","work_id":"a1f2574b-a899-4713-be60-c87ba332656c","shared_citers":6},{"title":"Brown, Miljan Martic, Shane Legg, and Dario Amodei","work_id":"95b596b9-4880-42ad-ac79-570a3a8a9dee","shared_citers":5},{"title":"Distilling the Knowledge in a Neural Network","work_id":"d927ab1f-17b8-4002-9d09-c3d55764fbad","shared_citers":5},{"title":"Fine-Tuning Language Models from Human Preferences","work_id":"4f54aad1-f3b6-404f-b9c7-e21ba0a33b99","shared_citers":5},{"title":"Measuring Mathematical Problem Solving With the MATH Dataset","work_id":"50652ac6-fb7c-4675-a2c2-159c241feb17","shared_citers":5}],"time_series":[{"n":2,"year":2023},{"n":5,"year":2024},{"n":53,"year":2026}],"dependency_candidates":[]},"error":null,"updated_at":"2026-05-14T10:49:25.401738+00:00"},"identity_refresh":{"job_type":"identity_refresh","status":"succeeded","result":{"items":[{"title":"Qwen3 Technical Report","outcome":"unchanged","work_id":"25a4e30c-1232-48e7-9925-02fa12ba7c9e","resolver":"local_arxiv","confidence":0.98,"old_work_id":"25a4e30c-1232-48e7-9925-02fa12ba7c9e"}],"counts":{"fixed":0,"merged":0,"unchanged":1,"quarantined":0,"needs_external_resolution":0},"errors":[],"attempted":1},"error":null,"updated_at":"2026-05-14T10:49:14.999448+00:00"},"role_polarity":{"job_type":"role_polarity","status":"succeeded","result":{"title":"Direct Preference Optimization: Your Language Model is Secretly a Reward Model","claims":[{"claim_text":"While large-scale unsupervised language models (LMs) learn broad world knowledge and some reasoning skills, achieving precise control of their behavior is difficult due to the completely unsupervised nature of their training. Existing methods for gaining such steerability collect human labels of the relative quality of model generations and fine-tune the unsupervised LM to align with these preferences, often with reinforcement learning from human feedback (RLHF). However, RLHF is a complex and often unstable procedure, first fitting a reward model that reflects the human preferences, and then ","claim_type":"abstract","evidence_strength":"source_metadata"},{"claim_text":"Logs Backward Slicing o3-mini, DeepSeek LLaMA3-70B PA, L-ACC AUCAD [198] Align Gen Fine-tuningMethod + Is- sue Logs Issue Alignment LLaMA3.1 Magicoder PA, LA, VF1 Zhong et al. [214] Full Gen Fine-tuning Java Method Logs Compact Model FTLLaMA-8B, Mistral GPT-4o, Claude PA, LA, F1 Specialized Logging Generation (Test Code, File-level, Configuration) File-Level [140] File Gen ICL Python File Multi Logs Pipeline Stages GPT-4o-mini - L-ACC, AOD Test-Code [151] Test Log ICL Test Method Assert/Log Test","claim_type":"background","confidence":0.9,"evidence_strength":"citation_context"},{"claim_text":"resulting in reward hacking. We address (1) by grounding the entire framework on diverse text sampled from a content-rich pre-training corpus. We ask the model to synthesize queries that draw from the text, and, in tasks for which ground truth exists, to construct queries whose answers are derivable from the text. We address (2) by using Direct Preference Optimization (DPO) [ 28], to train the LLM via contrastive signals from the response pairs. We use two methods to mitigate (3). First, we give","claim_type":"method","confidence":0.9,"evidence_strength":"citation_context"},{"claim_text":"pLogP, and QED), the LLM-based models consistently achieve much higher similarity to the source molecule, with SIM typically in the range of 0.48-0.59, whereas the non-LLM baselines remain in a much lower range 11 Table 6: Comparison with representative non-LLM baselines under a similarity-aware PMO objective. The main comparison uses a generation budget of 20. Method DRD2 GSK3β pLogP QED SR↑SIM↑RI↑ SR↑SIM↑RI↑ SR↑SIM↑RI↑ SR↑SIM↑RI↑ GraphGA (20) 30.80 0.210 12.86 100.000.077 15.48 100.000.154 11.","claim_type":"baseline","confidence":0.9,"evidence_strength":"citation_context"},{"claim_text":"Despite their impressive fluency, LLMs struggle with hallucination, generat- ing text that is plausible but factually ungrounded [17,19]. When extended to the visual domain, VLMs inherit this vulnerability and exhibit visual halluci- nation-confidently describing entities, attributes, or actions that are entirely absent from the visual input even if semantically plausible [15,29,38]. Conse- quently, as models evolve to process increasingly diverse sensory streams, main- taining strict factual gr","claim_type":"background","confidence":0.9,"evidence_strength":"citation_context"},{"claim_text":"where πθ(y|x, ϕ) denotes the policy induced by parametersθ when conditioned on textual contextϕ and queryx. For a task distributionDand rewardr, the natural joint objective is max θ, ϕ J(θ, ϕ) =E x∼D, y∼π θ(·|x,ϕ) \u0002 r(x, y) \u0003 .(2) Each factor admits many concrete optimizers. On the slow side,θ can be updated by SFT, preference optimization [43], or policy-gradient methods such as PPO [46] and GRPO [47], frequently under verifiable rewards [26]. On the fast side,ϕ can be updated by automated prom","claim_type":"method","confidence":0.9,"evidence_strength":"citation_context"},{"claim_text":"Commonsense Reasoning: We included PIQA[4], SIQA[63], HellaSwag[89], WinoGrande [62], ARC[11], OpenBookQA(OBQA)[50], and CommonsenseQA(CSQA)[ 75] to assess common sense reasoning. CSQA was exclusively tested using a 7-shot setup, while all other tests were conducted with a 0-shot configuration. Reading Comprehension: For reading comprehension, we report the 0-shot average on SQuAD[61], QuAC[8], and BoolQ[10]. Math: We report the average of the GSM8K[12] (8 shot), and MATH[28] (4 shot) benchmarks","claim_type":"dataset","confidence":0.9,"evidence_strength":"citation_context"}],"why_cited":"Pith tracks Direct Preference Optimization: Your Language Model is Secretly a Reward Model because it crossed a citation-hub threshold. Current citing contexts most often use it as background evidence (20 contexts).","role_counts":[{"n":20,"context_role":"background"},{"n":8,"context_role":"method"},{"n":1,"context_role":"baseline"},{"n":1,"context_role":"dataset"}]},"error":null,"updated_at":"2026-05-21T05:42:38.322629+00:00"},"summary_claims":{"job_type":"summary_claims","status":"succeeded","result":{"title":"Direct Preference Optimization: Your Language Model is Secretly a Reward Model","claims":[{"claim_text":"While large-scale unsupervised language models (LMs) learn broad world knowledge and some reasoning skills, achieving precise control of their behavior is difficult due to the completely unsupervised nature of their training. Existing methods for gaining such steerability collect human labels of the relative quality of model generations and fine-tune the unsupervised LM to align with these preferences, often with reinforcement learning from human feedback (RLHF). However, RLHF is a complex and often unstable procedure, first fitting a reward model that reflects the human preferences, and then ","claim_type":"abstract","evidence_strength":"source_metadata"}],"why_cited":"Pith tracks Direct Preference Optimization: Your Language Model is Secretly a Reward Model because it crossed a citation-hub threshold.","role_counts":[]},"error":null,"updated_at":"2026-05-14T10:49:12.714260+00:00"}},"summary":{"title":"Direct Preference Optimization: Your Language Model is Secretly a Reward Model","claims":[{"claim_text":"While large-scale unsupervised language models (LMs) learn broad world knowledge and some reasoning skills, achieving precise control of their behavior is difficult due to the completely unsupervised nature of their training. Existing methods for gaining such steerability collect human labels of the relative quality of model generations and fine-tune the unsupervised LM to align with these preferences, often with reinforcement learning from human feedback (RLHF). However, RLHF is a complex and often unstable procedure, first fitting a reward model that reflects the human preferences, and then ","claim_type":"abstract","evidence_strength":"source_metadata"}],"why_cited":"Pith tracks Direct Preference Optimization: Your Language Model is Secretly a Reward Model because it crossed a citation-hub threshold.","role_counts":[]},"graph":{"co_cited":[{"title":"Proximal Policy Optimization Algorithms","work_id":"240c67fe-d14d-4520-91c1-38a4e272ca19","shared_citers":16},{"title":"DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models","work_id":"c5006563-f3ec-438a-9e35-b7b484f34828","shared_citers":15},{"title":"Training language models to follow instructions with human feedback","work_id":"52aff42f-4fa9-4fcf-bdb3-1459b9bebf65","shared_citers":12},{"title":"Training Verifiers to Solve Math Word Problems","work_id":"acab1aa8-b4d6-40e0-a3ee-25341701dca2","shared_citers":12},{"title":"Qwen3 Technical Report","work_id":"25a4e30c-1232-48e7-9925-02fa12ba7c9e","shared_citers":11},{"title":"GPT-4 Technical Report","work_id":"b928e041-6991-4c08-8c81-0359e4097c7b","shared_citers":10},{"title":"The Llama 3 Herd of Models","work_id":"1549a635-88af-4ac1-acfe-51ae7bb53345","shared_citers":10},{"title":"Evaluating Large Language Models Trained on Code","work_id":"042493e9-b26f-4b4e-bbde-382072ca9b08","shared_citers":9},{"title":"Llama 2: Open Foundation and Fine-Tuned Chat Models","work_id":"68a5177f-d644-44c1-bd4f-4e5278c22f5d","shared_citers":8},{"title":"LLaMA: Open and Efficient Foundation Language Models","work_id":"c018fc23-6f3f-4035-9d02-28a2173b2b9d","shared_citers":8},{"title":"Constitutional AI: Harmlessness from AI Feedback","work_id":"faaaa4e0-2676-4fac-a0b4-99aef10d2095","shared_citers":7},{"title":"DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning","work_id":"e6b75ad5-2877-4168-97c8-710407094d20","shared_citers":7},{"title":"LoRA: Low-Rank Adaptation of Large Language Models","work_id":"0426219a-789e-4964-adc8-a04538510818","shared_citers":7},{"title":"Measuring Massive Multitask Language Understanding","work_id":"e87ec49a-544b-4ec8-8991-75298c64ff5e","shared_citers":7},{"title":"DAPO: An Open-Source LLM Reinforcement Learning System at Scale","work_id":"64019d00-0b11-4bbd-b173-b46c8fad0157","shared_citers":6},{"title":"Gemini: A Family of Highly Capable Multimodal Models","work_id":"83f7c85b-3f11-450f-ac0c-64d9745220b2","shared_citers":6},{"title":"Language Models are Few-Shot Learners","work_id":"214732c0-2edd-44a0-af9e-28184a2b8279","shared_citers":6},{"title":"Program Synthesis with Large Language Models","work_id":"fd241a05-03b9-4de2-9588-9d77ce176125","shared_citers":6},{"title":"Scaling Instruction-Finetuned Language Models","work_id":"8405abb1-7558-4fdf-af24-f4c52fa77a06","shared_citers":6},{"title":"Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback","work_id":"a1f2574b-a899-4713-be60-c87ba332656c","shared_citers":6},{"title":"Brown, Miljan Martic, Shane Legg, and Dario Amodei","work_id":"95b596b9-4880-42ad-ac79-570a3a8a9dee","shared_citers":5},{"title":"Distilling the Knowledge in a Neural Network","work_id":"d927ab1f-17b8-4002-9d09-c3d55764fbad","shared_citers":5},{"title":"Fine-Tuning Language Models from Human Preferences","work_id":"4f54aad1-f3b6-404f-b9c7-e21ba0a33b99","shared_citers":5},{"title":"Measuring Mathematical Problem Solving With the MATH Dataset","work_id":"50652ac6-fb7c-4675-a2c2-159c241feb17","shared_citers":5}],"time_series":[{"n":2,"year":2023},{"n":5,"year":2024},{"n":53,"year":2026}],"dependency_candidates":[]},"authors":[{"id":"52423bdb-17c9-48b5-a88d-611abbecf1fb","orcid":null,"display_name":"Archit Sharma","source":"manual","import_confidence":0.72},{"id":"379e406e-0cbc-4ede-b9dd-9a76a16a6da8","orcid":null,"display_name":"Chelsea Finn","source":"manual","import_confidence":0.72},{"id":"d649e132-c634-4aa5-9f58-83571eb7fc75","orcid":null,"display_name":"Christopher D. Manning","source":"manual","import_confidence":0.72},{"id":"bebec598-67af-4cdc-8623-d141039043ed","orcid":null,"display_name":"Eric Mitchell","source":"manual","import_confidence":0.72},{"id":"bafd1009-1b9f-4624-8ceb-ecf73ce7f982","orcid":null,"display_name":"Rafael Rafailov","source":"manual","import_confidence":0.72},{"id":"e00f90b7-aa2a-4902-8d30-19cdf0a496f9","orcid":null,"display_name":"Stefano Ermon","source":"manual","import_confidence":0.72}]}}