pith. machine review for the scientific record. sign in

arxiv: 2405.01470 · v1 · submitted 2024-05-02 · 💻 cs.CL

Recognition: no theorem link

WildChat: 1M ChatGPT Interaction Logs in the Wild

Wenting Zhao , Xiang Ren , Jack Hessel , Claire Cardie , Yejin Choi , Yuntian Deng
Authors on Pith no claims yet

Pith reviewed 2026-05-16 10:40 UTC · model grok-4.3

classification 💻 cs.CL
keywords WildChatChatGPT conversationsuser interaction logsopt-in datasetmultilingual promptstoxic use casesconversation corpusinstruction fine-tuning
0
0 comments X

The pith

A corpus of one million real ChatGPT conversations was assembled from users who opted in for free access.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The authors gave online users free ChatGPT access in return for consent to anonymously log their chat transcripts and request headers. This produced WildChat, a dataset of 1 million conversations totaling over 2.5 million turns. The collection is presented as having greater prompt diversity, more languages represented, and a broader set of potentially toxic examples than earlier public chat logs. The data also include location details and headers that support geographic and time-based breakdowns. Public release lets researchers examine actual usage patterns and fine-tune models on authentic user exchanges.

Core claim

WildChat is a corpus of 1 million user-ChatGPT conversations consisting of over 2.5 million interaction turns. It was compiled through an opt-in process where users received free access in exchange for consenting to the anonymous collection of their chat transcripts and request headers. The dataset offers the most diverse user prompts, contains the largest number of languages, and presents the richest variety of potentially toxic use-cases among available resources, while also including demographic information such as state, country, and hashed IP addresses for regional and temporal analysis.

What carries the argument

The opt-in consent collection process that built the WildChat corpus of timestamped ChatGPT transcripts, augmented with geographic and header metadata.

If this is right

  • Researchers gain the ability to analyze user behaviors across specific countries and time periods using the added location and timestamp data.
  • Instruction-following models can be fine-tuned on a broad range of authentic, real-world prompts drawn from the corpus.
  • Studies of potentially toxic interactions can draw on the largest captured variety of such cases for safety research.
  • Direct comparisons become possible between this dataset and smaller prior chat logs to quantify differences in prompt diversity and language coverage.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The release could encourage similar opt-in collections for other chatbots, creating comparable public resources across models.
  • Hashed IP data might allow researchers to study whether response quality or safety features vary by region without identifying individuals.
  • Fine-tuning experiments on the data could reveal whether exposure to toxic examples improves or harms model refusal behavior.
  • Temporal metadata opens the door to tracking shifts in user topics as the underlying model versions change over time.

Load-bearing premise

The opt-in consent process with a free-access incentive yields a representative sample of ChatGPT users without major selection bias.

What would settle it

A side-by-side comparison of conversation topics, language distribution, or toxicity rates between WildChat and a random sample of actual ChatGPT logs that shows large systematic differences would indicate the collection method introduced bias.

read the original abstract

Chatbots such as GPT-4 and ChatGPT are now serving millions of users. Despite their widespread use, there remains a lack of public datasets showcasing how these tools are used by a population of users in practice. To bridge this gap, we offered free access to ChatGPT for online users in exchange for their affirmative, consensual opt-in to anonymously collect their chat transcripts and request headers. From this, we compiled WildChat, a corpus of 1 million user-ChatGPT conversations, which consists of over 2.5 million interaction turns. We compare WildChat with other popular user-chatbot interaction datasets, and find that our dataset offers the most diverse user prompts, contains the largest number of languages, and presents the richest variety of potentially toxic use-cases for researchers to study. In addition to timestamped chat transcripts, we enrich the dataset with demographic data, including state, country, and hashed IP addresses, alongside request headers. This augmentation allows for more detailed analysis of user behaviors across different geographical regions and temporal dimensions. Finally, because it captures a broad range of use cases, we demonstrate the dataset's potential utility in fine-tuning instruction-following models. WildChat is released at https://wildchat.allen.ai under AI2 ImpACT Licenses.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The paper presents WildChat, a corpus of 1 million user-ChatGPT conversations (over 2.5 million turns) collected by offering free ChatGPT access in exchange for opt-in consent to share transcripts and request headers. It claims superiority over prior datasets in prompt diversity, language coverage, and variety of toxic use-cases, augments the data with country/state/hashed-IP demographics, and demonstrates utility for fine-tuning instruction-following models.

Significance. If the diversity, language, and toxicity claims hold after bias correction, WildChat would be a significant public resource as the largest released corpus of real-world ChatGPT interactions, supporting research on usage patterns, multilingual behavior, toxicity, and model alignment.

major comments (3)
  1. [Data Collection] Data Collection section: the opt-in free-access incentive structure selects for non-subscription users willing to share logs; no quantitative comparison to known ChatGPT user demographics, no inverse-probability weighting, and no sensitivity analysis are reported to show that the diversity/language/toxicity rankings survive plausible re-weighting.
  2. [Comparisons] Comparisons section (and abstract claims): the metrics establishing 'most diverse user prompts' and 'largest number of languages' are not detailed with explicit formulas or controls for sampling bias, so the superiority statements rest on unverified assertions relative to prior datasets.
  3. [Toxicity Analysis] Toxicity analysis: the process for labeling 'potentially toxic use-cases' (e.g., tools, thresholds, or human annotation protocol) is not described, undermining the claim of 'richest variety' and preventing independent verification.
minor comments (2)
  1. [Abstract] Abstract: specify concrete numbers (e.g., exact language count or diversity metric values) rather than qualitative superlatives.
  2. [Release] Dataset release: clarify the exact terms of the AI2 ImpACT Licenses and any usage restrictions in the main text.

Simulated Author's Rebuttal

3 responses · 1 unresolved

We thank the referee for the constructive and detailed feedback. We address each major comment below, indicating where revisions will be made to strengthen the manuscript while being transparent about inherent limitations of the data collection approach.

read point-by-point responses

  1. Referee: [Data Collection] Data Collection section: the opt-in free-access incentive structure selects for non-subscription users willing to share logs; no quantitative comparison to known ChatGPT user demographics, no inverse-probability weighting, and no sensitivity analysis are reported to show that the diversity/language/toxicity rankings survive plausible re-weighting.

    Authors: We agree that the opt-in free-access model introduces selection bias toward non-subscribing users willing to share logs. This was a deliberate ethical choice to obtain affirmative consent. We lack access to proprietary ChatGPT user demographics, so a direct quantitative comparison, inverse-probability weighting, or sensitivity analysis is not feasible. We will add a limitations subsection explicitly discussing these biases and noting that the released geographic and header metadata allow downstream researchers to perform their own re-weighting or sensitivity checks. revision: partial

  2. Referee: [Comparisons] Comparisons section (and abstract claims): the metrics establishing 'most diverse user prompts' and 'largest number of languages' are not detailed with explicit formulas or controls for sampling bias, so the superiority statements rest on unverified assertions relative to prior datasets.

    Authors: We will expand the Comparisons section to include explicit formulas for prompt diversity (unique normalized prompts and type-token ratio) and language coverage (language identification library and detection thresholds). We will also add a paragraph addressing sampling bias relative to prior datasets and how it may affect the reported rankings, while retaining the raw comparative counts that support broader coverage. revision: yes

  3. Referee: [Toxicity Analysis] Toxicity analysis: the process for labeling 'potentially toxic use-cases' (e.g., tools, thresholds, or human annotation protocol) is not described, undermining the claim of 'richest variety' and preventing independent verification.

    Authors: We apologize for the missing description. Toxicity labeling combined the Perspective API with fixed score thresholds and targeted manual review of borderline cases. We will insert a dedicated subsection detailing the exact tools, thresholds, annotation guidelines, and any agreement statistics to enable verification and to substantiate the variety claim. revision: yes

standing simulated objections not resolved
  • Quantitative comparison to known ChatGPT user demographics (proprietary data unavailable to the authors)

Circularity Check

0 steps flagged

No circularity: observational dataset collection with direct empirical comparisons

full rationale

The paper contains no derivations, equations, fitted parameters, or predictive models. It describes an opt-in data collection process, releases the resulting logs, and performs straightforward empirical comparisons of diversity metrics against prior datasets. All claims reduce directly to the collected data without any self-referential reduction or load-bearing self-citation chains. The work is fully self-contained as an observational release.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the validity of the opt-in consent mechanism and the assumption that participating users yield representative interaction data; no free parameters or invented entities are introduced.

axioms (1)
  • domain assumption Opt-in users who receive free access provide interaction data representative of broader ChatGPT usage without substantial selection bias
    Invoked to support claims of diversity and generalizability in the abstract.

pith-pipeline@v0.9.0 · 5535 in / 1226 out tokens · 40461 ms · 2026-05-16T10:40:02.109654+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 18 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Instruction Tuning Changes How Upstream State Conditions Late Readout: A Cross-Patching Diagnostic

    cs.LG 2026-05 unverdicted novelty 7.0

    Instruction tuning makes late-layer computation depend more on the model's own post-trained upstream state than on base-model upstream state, producing a consistent +1.68 logit interaction effect across five model families.

  2. The Partial Testimony of Logs: Evaluation of Language Model Generation under Confounded Model Choice

    cs.LG 2026-05 unverdicted novelty 7.0

    An identification theorem shows that a randomized experiment and simulator together recover causal model values from confounded logs, with logs used only afterward to reduce estimation error.

  3. CacheFlow: Efficient LLM Serving with 3D-Parallel KV Cache Restoration

    cs.DC 2026-04 unverdicted novelty 7.0

    CacheFlow cuts TTFT by 10-62% in batched LLM serving via 3D-parallel KV cache restoration and a two-pointer scheduler that overlaps recompute and I/O.

  4. Beyond Semantic Manipulation: Token-Space Attacks on Reward Models

    cs.LG 2026-04 unverdicted novelty 7.0

    TOMPA performs black-box adversarial optimization in token space to discover non-linguistic patterns that nearly double the reward scores of GPT-5 answers on Skywork-Reward-V2 while producing gibberish text.

  5. Analytical Provisioning for Attention-FFN Disaggregated LLM Serving under Stochastic Workloads

    cs.LG 2026-01 unverdicted novelty 7.0

    A renewal-reward analysis yields a closed-form mean-field rule for the optimal Attention/FFN provisioning ratio in disaggregated LLM serving that accounts for stochastic KV-cache growth and matches simulation optima w...

  6. Grounded Continuation: A Linear-Time Runtime Verifier for LLM Conversations

    cs.AI 2026-05 conditional novelty 6.0

    A hybrid LLM-symbolic verifier maintains a dependency graph over conversation turns classified into eight formal update operations, enabling linear-time groundedness checks and precise retraction propagation with a co...

  7. Enabling Performant and Flexible Model-Internal Observability for LLM Inference

    cs.LG 2026-05 unverdicted novelty 6.0

    DMI-Lib delivers 0.4-6.8% overhead for offline batch LLM inference and ~6% for moderate online serving while exposing rich internal signals across backends, cutting latency overhead 2-15x versus prior observability baselines.

  8. Annotations Mitigate Post-Training Mode Collapse

    cs.CL 2026-05 unverdicted novelty 6.0

    Annotation-anchored training reduces semantic diversity collapse in post-trained language models by a factor of six compared to standard supervised fine-tuning while preserving instruction-following and improving with scale.

  9. Estimating the Black-box LLM Uncertainty with Distribution-Aligned Adversarial Distillation

    cs.CL 2026-05 unverdicted novelty 6.0

    DisAAD trains a 1%-sized proxy model via adversarial distillation to quantify uncertainty in black-box LLMs by aligning with their output distributions.

  10. Chain of Risk: Safety Failures in Large Reasoning Models and Mitigation via Adaptive Multi-Principle Steering

    cs.AI 2026-05 unverdicted novelty 6.0

    Reasoning traces in large reasoning models expose safety failures missed by final-answer checks, and adaptive multi-principle steering reduces unsafe content in both traces and answers while preserving task performance.

  11. Stayin' Aligned Over Time: Towards Longitudinal Human-LLM Alignment via Contextual Reflection and Privacy-Preserving Behavioral Data

    cs.HC 2026-05 unverdicted novelty 6.0

    A methodological framework and browser system BITE for collecting evolving user preferences on LLM outputs through context-triggered reflections and privacy-preserving data over time.

  12. Rethinking Network Topologies for Cost-Effective Mixture-of-Experts LLM Serving

    cs.NI 2026-04 unverdicted novelty 6.0

    Switchless topologies such as 3D full-mesh are 20.6-56.2% more cost-effective than scale-up networks for MoE LLM serving, with current link bandwidths over-provisioned by up to 27%.

  13. Length Value Model: Scalable Value Pretraining for Token-Level Length Modeling

    cs.CL 2026-04 unverdicted novelty 6.0

    LenVM models token-level remaining generation length as a bounded discounted value function derived from constant negative per-token rewards, providing a scalable proxy for generation horizon.

  14. A paradox of AI fluency

    cs.CL 2026-04 unverdicted novelty 6.0

    Fluent AI users adopt an active, iterative collaboration mode that produces more visible failures but better recovery and success on hard tasks, whereas novices experience more invisible failures from passive use.

  15. From Searchable to Non-Searchable: Generative AI and Information Diversity in Online Information Seeking

    cs.HC 2026-04 unverdicted novelty 6.0

    ChatGPT expands the diversity of user questions (80% non-searchable) but delivers less diverse responses than Google for comparable queries, creating a feedback loop that may constrain information exposure.

  16. Language Model Goal Selection Differs from Humans' in a Self-Directed Learning Task

    cs.CL 2026-02 unverdicted novelty 6.0

    LLMs diverge from human goal selection in self-directed learning by exploiting single solutions with low variability across instances.

  17. Quantifying the Utility of User Simulators for Building Collaborative LLM Assistants

    cs.CL 2026-05 unverdicted novelty 5.0

    Fine-tuned simulators grounded in real human data produce LLM assistants that win more often against real users than those trained against role-playing simulators.

  18. Same Voice, Different Lab: On the Homogenization of Frontier LLM Personalities

    cs.HC 2026-03 unverdicted novelty 5.0

    Frontier LLMs homogenize toward systematic and analytical personalities, suppressing emotional traits like remorseful or sycophantic, indicating an implicit consensus on optimal assistant behavior.

Reference graph

Works this paper leans on

54 extracted references · 54 canonical work pages · cited by 18 Pith papers · 1 internal anchor

  1. [1]

    Model card and evaluations for claude models, Jul 2023

    Anthropic. Model card and evaluations for claude models, Jul 2023. URL https://www-cdn.anthropic.com/bd2a28d2535bfb0494cc8e2a3bf135d2e7523226/Model-Card-Claude-2.pdf

    work page 2023

  2. [2]

    Bowman, Zac Hatfield-Dodds, Ben Mann, Dario Amodei, Nicholas Joseph, Sam McCandlish, Tom Brown, and Jared Kaplan

    Yuntao Bai, Saurav Kadavath, Sandipan Kundu, Amanda Askell, Jackson Kernion, Andy Jones, Anna Chen, Anna Goldie, Azalia Mirhoseini, Cameron McKinnon, Carol Chen, Catherine Olsson, Christopher Olah, Danny Hernandez, Dawn Drain, Deep Ganguli, Dustin Li, Eli Tran-Johnson, Ethan Perez, Jamie Kerr, Jared Mueller, Jeffrey Ladish, Joshua Landau, Kamal Ndousse, K...

    work page 2022

  3. [3]

    Free dolly: Introducing the world’s first truly open instruction-tuned llm, 2023

    Mike Conover, Matt Hayes, Ankit Mathur, Xiangrui Meng, Jianwei Xie, Jun Wan, Sam Shah, Ali Ghodsi, Patrick Wendell, Matei Zaharia, et al. Free dolly: Introducing the world’s first truly open instruction-tuned llm, 2023

    work page 2023

  4. [4]

    Bard: A conversational ai tool by google, 2023

    Google. Bard: A conversational ai tool by google, 2023. URL https://bard.google.com/. Accessed: Sep 27, 2023

    work page 2023

  5. [5]

    Detoxify

    Laura Hanu and Unitary team . Detoxify. Github. https://github.com/unitaryai/detoxify, 2020

    work page 2020

  6. [6]

    Deberta: Decoding-enhanced bert with disentangled attention, 2021

    Pengcheng He, Xiaodong Liu, Jianfeng Gao, and Weizhu Chen. Deberta: Decoding-enhanced bert with disentangled attention, 2021

    work page 2021

  7. [8]

    Wildbench: Benchmarking llms with challenging tasks from real users in the wild, 2024

    Bill Yuchen Lin, Khyathi Chandu, Faeze Brahman, Yuntian Deng, Abhilasha Ravichander, Valentina Pyatkin, Ronan Le Bras, and Yejin Choi. Wildbench: Benchmarking llms with challenging tasks from real users in the wild, 2024. URL https://huggingface.co/spaces/allenai/WildBench

    work page 2024

  8. [9]

    Introducing the new bing, 2023

    Microsoft. Introducing the new bing, 2023. URL https://www.bing.com/new. Accessed: Sep 27, 2023

    work page 2023

  9. [10]

    GPT-4 Technical Report

    OpenAI. Gpt-4 technical report. ArXiv, abs/2303.08774, 2023. URL https://api.semanticscholar.org/CorpusID:257532815

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  10. [11]

    Training language models to follow instructions with human feedback

    Long Ouyang, Jeffrey Wu, Xu Jiang, Diogo Almeida, Carroll Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, John Schulman, Jacob Hilton, Fraser Kelton, Luke Miller, Maddie Simens, Amanda Askell, Peter Welinder, Paul F Christiano, Jan Leike, and Ryan Lowe. Training language models to follow instructions with human feedbac...

    work page 2022

  11. [12]

    Direct preference optimization: Your language model is secretly a reward model

    Rafael Rafailov, Archit Sharma, Eric Mitchell, Christopher D Manning, Stefano Ermon, and Chelsea Finn. Direct preference optimization: Your language model is secretly a reward model. In Thirty-seventh Conference on Neural Information Processing Systems, 2023. URL https://openreview.net/forum?id=HPuSIXJaa9

    work page 2023

  12. [13]

    Is reinforcement learning (not) for natural language processing: Benchmarks, baselines, and building blocks for natural language policy optimization

    Rajkumar Ramamurthy, Prithviraj Ammanabrolu, Kiant \'e Brantley, Jack Hessel, Rafet Sifa, Christian Bauckhage, Hannaneh Hajishirzi, and Yejin Choi. Is reinforcement learning (not) for natural language processing: Benchmarks, baselines, and building blocks for natural language policy optimization. In The Eleventh International Conference on Learning Repres...

    work page 2023

  13. [14]

    Peters, Abhilasha Ravichander, Kyle Richardson, Zejiang Shen, Emma Strubell, Nishant Subramani, Oyvind Tafjord, Pete Walsh, Luke Zettlemoyer, Noah A

    Luca Soldaini, Rodney Kinney, Akshita Bhagia, Dustin Schwenk, David Atkinson, Russell Authur, Ben Bogin, Khyathi Chandu, Jennifer Dumas, Yanai Elazar, Valentin Hofmann, Ananya Harsh Jha, Sachin Kumar, Li Lucy, Xinxi Lyu, Nathan Lambert, Ian Magnusson, Jacob Morrison, Niklas Muennighoff, Aakanksha Naik, Crystal Nam, Matthew E. Peters, Abhilasha Ravichander...

    work page arXiv 2024

  14. [15]

    Learning to summarize with human feedback

    Nisan Stiennon, Long Ouyang, Jeffrey Wu, Daniel Ziegler, Ryan Lowe, Chelsea Voss, Alec Radford, Dario Amodei, and Paul F Christiano. Learning to summarize with human feedback. In H. Larochelle, M. Ranzato, R. Hadsell, M.F. Balcan, and H. Lin (eds.), Advances in Neural Information Processing Systems, volume 33, pp.\ 3008--3021. Curran Associates, Inc., 202...

    work page 2020

  15. [16]

    Hashimoto

    Rohan Taori, Ishaan Gulrajani, Tianyi Zhang, Yann Dubois, Xuechen Li, Carlos Guestrin, Percy Liang, and Tatsunori B. Hashimoto. Stanford alpaca: An instruction-following llama model. https://github.com/tatsu-lab/stanford_alpaca, 2023

    work page 2023

  16. [17]

    Llama 2: Open foundation and fine-tuned chat models, 2023

    Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, Dan Bikel, Lukas Blecher, Cristian Canton Ferrer, Moya Chen, Guillem Cucurull, David Esiobu, Jude Fernandes, Jeremy Fu, Wenyin Fu, Brian Fuller, Cynthia Gao, Vedanuj Goswami, Naman Goyal, Anthony Harts...

    work page 2023

  17. [18]

    Visualizing data using t-sne

    Laurens Van der Maaten and Geoffrey Hinton. Visualizing data using t-sne. Journal of machine learning research, 9 0 (11), 2008

    work page 2008

  18. [19]

    Chatgpt loses users for first time, shaking faith in ai revolution, Jul 2023

    Gerrit De Vynck. Chatgpt loses users for first time, shaking faith in ai revolution, Jul 2023. URL https://www.washingtonpost.com/technology/2023/07/07/chatgpt-users-decline-future-ai-openai/. Accessed: Sep 27, 2023

    work page 2023

  19. [20]

    Smith, Hannaneh Hajishirzi, and Daniel Khashabi

    Yizhong Wang, Swaroop Mishra, Pegah Alipoormolabashi, Yeganeh Kordi, Amirreza Mirzaei, Anjana Arunkumar, Arjun Ashok, Arut Selvan Dhanasekaran, Atharva Naik, David Stap, Eshaan Pathak, Giannis Karamanolakis, Haizhi Gary Lai, Ishan Purohit, Ishani Mondal, Jacob Anderson, Kirby Kuznia, Krima Doshi, Maitreya Patel, Kuntal Kumar Pal, Mehrad Moradshahi, Mihir ...

    work page 2022

  20. [21]

    Smith, Mari Ostendorf, and Hannaneh Hajishirzi

    Zeqiu Wu, Yushi Hu, Weijia Shi, Nouha Dziri, Alane Suhr, Prithviraj Ammanabrolu, Noah A. Smith, Mari Ostendorf, and Hannaneh Hajishirzi. Fine-grained human feedback gives better rewards for language model training. In Thirty-seventh Conference on Neural Information Processing Systems, 2023. URL https://openreview.net/forum?id=CSbGXyCswu

    work page 2023

  21. [22]

    P Xing, Hao Zhang, Joseph E

    Lianmin Zheng, Wei-Lin Chiang, Ying Sheng, Siyuan Zhuang, Zhanghao Wu, Yonghao Zhuang, Zi Lin, Zhuohan Li, Dacheng Li, Eric. P Xing, Hao Zhang, Joseph E. Gonzalez, and Ion Stoica. Judging llm-as-a-judge with mt-bench and chatbot arena, 2023

    work page 2023

  22. [23]

    Gonzalez, Ion Stoica, and Hao Zhang

    Lianmin Zheng, Wei-Lin Chiang, Ying Sheng, Tianle Li, Siyuan Zhuang, Zhanghao Wu, Yonghao Zhuang, Zhuohan Li, Zi Lin, Eric Xing, Joseph E. Gonzalez, Ion Stoica, and Hao Zhang. Lmsys-chat-1m: A large-scale real-world LLM conversation dataset. In The Twelfth International Conference on Learning Representations, 2024. URL https://openreview.net/forum?id=BOfDKxfwt0

    work page 2024

  23. [24]

    Lima: Less is more for alignment, 2023

    Chunting Zhou, Pengfei Liu, Puxin Xu, Srini Iyer, Jiao Sun, Yuning Mao, Xuezhe Ma, Avia Efrat, Ping Yu, Lili Yu, Susan Zhang, Gargi Ghosh, Mike Lewis, Luke Zettlemoyer, and Omer Levy. Lima: Less is more for alignment, 2023

    work page 2023

  24. [25]

    Scaling Learning Algorithms Towards

    Bengio, Yoshua and LeCun, Yann , booktitle =. Scaling Learning Algorithms Towards

    work page

  25. [26]

    and Osindero, Simon and Teh, Yee Whye , journal =

    Hinton, Geoffrey E. and Osindero, Simon and Teh, Yee Whye , journal =. A Fast Learning Algorithm for Deep Belief Nets , volume =

    work page

  26. [27]

    2016 , publisher=

    Deep learning , author=. 2016 , publisher=

    work page 2016

  27. [28]

    2023 , eprint=

    Llama 2: Open Foundation and Fine-Tuned Chat Models , author=. 2023 , eprint=

    work page 2023

  28. [29]

    2023 , eprint=

    Judging LLM-as-a-judge with MT-Bench and Chatbot Arena , author=. 2023 , eprint=

    work page 2023

  29. [30]

    2023 , eprint=

    LIMA: Less Is More for Alignment , author=. 2023 , eprint=

    work page 2023

  30. [31]

    arXiv preprint arXiv:2306.04707 , year=

    Improving Open Language Models by Learning from Organic Interactions , author=. arXiv preprint arXiv:2306.04707 , year=

    work page arXiv

  31. [32]

    Hashimoto , title =

    Rohan Taori and Ishaan Gulrajani and Tianyi Zhang and Yann Dubois and Xuechen Li and Carlos Guestrin and Percy Liang and Tatsunori B. Hashimoto , title =. GitHub repository , howpublished =. 2023 , publisher =

    work page 2023

  32. [33]

    arXiv preprint arXiv:2304.07327 , year=

    OpenAssistant Conversations--Democratizing Large Language Model Alignment , author=. arXiv preprint arXiv:2304.07327 , year=

    work page arXiv

  33. [34]

    Free dolly: Introducing the world’s first truly open instruction-tuned llm , author=

    work page

  34. [35]

    , author=

    Visualizing data using t-SNE. , author=. Journal of machine learning research , volume=

    work page

  35. [36]

    Bender, Timnit Gebru, Angelina McMillan-Major, and Shmargaret Shmitchell

    Bender, Emily M. and Gebru, Timnit and McMillan-Major, Angelina and Shmitchell, Shmargaret , title =. Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency , pages =. 2021 , isbn =. doi:10.1145/3442188.3445922 , abstract =

    work page doi:10.1145/3442188.3445922 2021

  36. [37]

    ArXiv , year=

    GPT-4 Technical Report , author=. ArXiv , year=

    work page

  37. [38]

    WildBench: Benchmarking LLMs with Challenging Tasks from Real Users in the Wild , author =

    work page

  38. [39]

    Dense Passage Retrieval for Open-Domain Question Answering

    Karpukhin, Vladimir and Oguz, Barlas and Min, Sewon and Lewis, Patrick and Wu, Ledell and Edunov, Sergey and Chen, Danqi and Yih, Wen-tau. Dense Passage Retrieval for Open-Domain Question Answering. Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP). 2020. doi:10.18653/v1/2020.emnlp-main.550

    work page doi:10.18653/v1/2020.emnlp-main.550 2020

  39. [40]

    2021 , eprint=

    DeBERTa: Decoding-enhanced BERT with Disentangled Attention , author=. 2021 , eprint=

    work page 2021

  40. [41]

    2022 , eprint=

    Super-NaturalInstructions: Generalization via Declarative Instructions on 1600+ NLP Tasks , author=. 2022 , eprint=

    work page 2022

  41. [42]

    2022 , eprint=

    Constitutional AI: Harmlessness from AI Feedback , author=. 2022 , eprint=

    work page 2022

  42. [43]

    Thirty-seventh Conference on Neural Information Processing Systems , year=

    Fine-Grained Human Feedback Gives Better Rewards for Language Model Training , author=. Thirty-seventh Conference on Neural Information Processing Systems , year=

    work page

  43. [44]

    Gonzalez and Ion Stoica and Hao Zhang , booktitle=

    Lianmin Zheng and Wei-Lin Chiang and Ying Sheng and Tianle Li and Siyuan Zhuang and Zhanghao Wu and Yonghao Zhuang and Zhuohan Li and Zi Lin and Eric Xing and Joseph E. Gonzalez and Ion Stoica and Hao Zhang , booktitle=. LMSYS-Chat-1M: A Large-Scale Real-World. 2024 , url=

    work page 2024

  44. [45]

    Thirty-seventh Conference on Neural Information Processing Systems , year=

    Direct Preference Optimization: Your Language Model is Secretly a Reward Model , author=. Thirty-seventh Conference on Neural Information Processing Systems , year=

    work page

  45. [46]

    Peters and Abhilasha Ravichander and Kyle Richardson and Zejiang Shen and Emma Strubell and Nishant Subramani and Oyvind Tafjord and Pete Walsh and Luke Zettlemoyer and Noah A

    Luca Soldaini and Rodney Kinney and Akshita Bhagia and Dustin Schwenk and David Atkinson and Russell Authur and Ben Bogin and Khyathi Chandu and Jennifer Dumas and Yanai Elazar and Valentin Hofmann and Ananya Harsh Jha and Sachin Kumar and Li Lucy and Xinxi Lyu and Nathan Lambert and Ian Magnusson and Jacob Morrison and Niklas Muennighoff and Aakanksha Na...

    work page

  46. [47]

    2023 , month =

    Anthropic , title =. 2023 , month =

    work page 2023

  47. [48]

    2024 , month =

    Anthropic , title =. 2024 , month =

    work page 2024

  48. [49]

    2023 , url =

    Google , title =. 2023 , url =

    work page 2023

  49. [50]

    2023 , url =

    Microsoft , title =. 2023 , url =

    work page 2023

  50. [51]

    2023 , month =

    Microsoft , title =. 2023 , month =

    work page 2023

  51. [52]

    The Washington Post , url =

    Gerrit De Vynck , title =. The Washington Post , url =. 2023 , month =

    work page 2023

  52. [53]

    Learning to summarize with human feedback , url =

    Stiennon, Nisan and Ouyang, Long and Wu, Jeffrey and Ziegler, Daniel and Lowe, Ryan and Voss, Chelsea and Radford, Alec and Amodei, Dario and Christiano, Paul F , booktitle =. Learning to summarize with human feedback , url =

    work page

  53. [54]

    Training language models to follow instructions with human feedback , url =

    Ouyang, Long and Wu, Jeffrey and Jiang, Xu and Almeida, Diogo and Wainwright, Carroll and Mishkin, Pamela and Zhang, Chong and Agarwal, Sandhini and Slama, Katarina and Ray, Alex and Schulman, John and Hilton, Jacob and Kelton, Fraser and Miller, Luke and Simens, Maddie and Askell, Amanda and Welinder, Peter and Christiano, Paul F and Leike, Jan and Lowe,...

    work page

  54. [55]

    The Eleventh International Conference on Learning Representations , year=

    Is Reinforcement Learning (Not) for Natural Language Processing: Benchmarks, Baselines, and Building Blocks for Natural Language Policy Optimization , author=. The Eleventh International Conference on Learning Representations , year=

    work page