WildChat: 1M ChatGPT Interaction Logs in the Wild
By Wenting Zhao, Xiang Ren, J. Hessel, Claire Cardie, Yejin Choi, Yuntian Deng. 2024
617 citations recorded by the source corpus.
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.
Subfields: Artificial Intelligence, Evaluation & Benchmarks, Dialogue & Conversational AI
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Artificial Intelligence
- Yejin Choi
- Ronan Le Bras
- Bill Yuchen Lin
- Mohit Bansal
- Abhilasha Ravichander
- Nouha Dziri
- Hannaneh Hajishirzi
- Radha Poovendran
In Evaluation & Benchmarks
- Ronan Le Bras
- Yejin Choi
- Bill Yuchen Lin
- Abhilasha Ravichander
- Radha Poovendran
- Nouha Dziri
- Hannaneh Hajishirzi
- Mohit Bansal
Citations in the corpus
- TÜLU 3: Pushing Frontiers in Open Language Model Post-Training (2024, neutral): “We pick datasets that can promote diversity, including: WildChat [Zhao et al., 2024], which is a large source of real-user interaction with models; Open Assistant [Köpf et al., 2024], which is created by volunteer workers for general [Rajani et al., 2023], which is annotated by expert workers for a…”
- OpenAI o1 System Card (2024, neutral): “• (Toxic) WildChat [16]: Toxic conversations from WildChat, which is a public corpus of 1M ChatGPT conversations labeled with ModAPI scores.”
- Instruction Tuning for Large Language Models: A Survey (2023, neutral): “Prompts, which are sometimes named task templates , are functions that map a data instance in a conventional NLP task (e.g., question answering, text classification) to a natural language input-output pair.”
- SimPO: Simple Preference Optimization with a Reference-Free Reward (2024, neutral): “The dataset used in this work, UltraFeedback [23], primarily focuses on helpfulness, and future research may consider a more comprehensive study utilizing larger-scale preference datasets [39, 90] and evaluation benchmarks [76] that place a strong emphasis on safety aspects.”
- HealthBench: Evaluating Large Language Models Towards Improved Human Health (2025, neutral): “These user-model interactions are highly varied and idiosyncratic, and reflect individual user styles in the same way that two people might send emails or text messages in very different styles (Zheng et al., 2024; Zhao et al., 2024).”
- Improving Alignment and Robustness with Circuit Breakers (2024, neutral): “Following the methodology outlined in [5], we construct an over-refusal evaluation using the WildChat dataset [74].”
- WildBench: Benchmarking LLMs with Challenging Tasks from Real Users in the Wild (2024, methodology): “The data used in W ILD B ENCH is sourced from the WildChat dataset [37], which reflects the demographic distribution of its users.”
- MixEval: Deriving Wisdom of the Crowd from LLM Benchmark Mixtures (2024, neutral): “ShareGPT, grounded in 100 million 4 active users of ChatGPT by mid-2023, contrasts with WildChat [38], Chatbot Arena Conversations [6], and LMSYS-Chat-1M [39], which have user bases of 0.2 million, 0.13 million, and 0.21 million, respectively.”
- Magpie: Alignment Data Synthesis from Scratch by Prompting Aligned LLMs with Nothing (2024, methodology): “To create datasets for alignment, previous studies have collected human interactions with LLMs [14, 64, 65, 66, 26].”
- Scaling Synthetic Data Creation with 1,000,000,000 Personas (2024, background): “By scaling this process with 1 billion personas in Persona Hub, we can easily obtain a vast array of knowledge-rich texts that cover almost any topic across various levels of granularity.”
- CUEing User Simulators: Calibrated User Embeddings for Multi-Turn Benchmarking (2026, unknown): “We use English multi-turn conversations from 24 DialogStudio datasets (Zhang et al., 2024a), LMSYS-Chat-1M (Zheng et al., 2024), and WildChat-1M (Zhao et al., 2024), balancing by capping each dataset at 10k samples.”