README.md

metadata

language:
  - en
license: cc-by-4.0
size_categories:
  - 1K<n<10K
pretty_name: WildChat-2k-TypeTopic

WildChat-2k-TypeTopic: The Manually Curated Edition

Dataset Description

WildChat-2k-TypeTopic is a manually curated subset of 1,880 real-world user prompts from the WildChat dataset, featuring annotations for both task type (e.g. knowledge recall, problem solving, creative, lists) and topic category (e.g. personal assistance, math, ai, household)

Why this dataset?

Suppose you want to answer a research question such as "What kind of user prompt does the LLM like doing most?" or "What is the implicit utility function of the LLM for answering different user prompts?" or "What kind of user prompts do models bail on"? The first step is to find a dataset of user prompts.

WildChat-1M is the most frequently used dataset of user prompts to LLMs; unfortunately everyone who has ever looked into it knows it is full of nonsensical prompts, typos, non-English, NSFW stuff, and other noise; and that the distribution of prompts that users ask for is very detailed in some domains (e.g. creative writing) and very sparse in others.

WildChat-2k-TypeTopic is a curated subset of single-message user prompts, constructed as follows:

1. Filter out (using an LLM filter) prompts that:

• are not in English
• are not meaningful tasks (e.g., random character strings, “hello”)
• are incomplete (e.g., “Fix this code” with no code provided)
• are infeasible for text-only LLMs (e.g., “Describe a time when you worked in a team”, “an image of a cat”)
• are clearly part of multi-turn conversations (e.g., text-based game setups)
• are more than 800 characters long

2. Deduplicate using text-embedding-3-large embeddings.
3. Classify into 16 task types and 25 topic categories, then subsample ~2000 tasks to preserve representation of all types and categories.
4. Manual review to remove anything problematic according to the described criteria.

Key Features

• 1,880 annotated prompts from real user interactions
• 15 task type categories (e.g., creative, coding, explanation, problem_solving)
• 24 topic categories (e.g., programming_other, creative_writing, personal_assistance)
• Short prompts: 12-800 characters (median: 116)
• Quality filtered: All entries are coherent English prompts, as opposed to WildChat

Dataset Structure

Data Format

The dataset is provided in JSONL format (newline-delimited JSON), with each entry containing:

{
  "id": "wildchat2k_0003",
  "text": "I want to learn how to understand and speak spanish, can you use the pareto principle, which identifies 20% of the topic that will yield 80% of the desired results, to create a learning plan for me?",
  "type": "planning_design",
  "topic": "languages"
}

Fields

• id (string): Unique identifier
• text (string): The user prompt/query
• type (string): Task classification (15 categories)
• topic (string): Subject matter classification (24 categories)

Task Types (15 categories)

Task Type	Count	%	Description
knowledge_recall	351	18.67%	Factual questions and information retrieval
creative	320	17.02%	Creative writing and content generation
explanation	305	16.22%	Requests for explanations and teaching
problem_solving	123	6.54%	Mathematical and logical problems
lists	123	6.54%	List generation tasks
rewriting	115	6.12%	Text rewriting and paraphrasing
coding	93	4.95%	Programming and code generation
analysis	86	4.57%	Analytical tasks
messaging	84	4.47%	Email and message writing
planning_design	68	3.62%	Planning and design tasks
translation	52	2.77%	Translation requests
summarization	51	2.71%	Summary generation
roleplay	44	2.34%	Roleplay and character simulation
decision_making	37	1.97%	Decision support tasks
evaluation	28	1.49%	Evaluation and assessment

Topic Categories (24 categories)

Topic	Count	%	Description
personal_assistance	227	12.07%	Personal productivity and communication
creative_writing	196	10.43%	Fiction, stories, creative content
programming_other	155	8.24%	Programming and software development
popular_culture	137	7.29%	Entertainment, media, celebrities
technology_other	106	5.64%	General technology topics
languages	105	5.59%	Language learning and linguistics
gaming	101	5.37%	Video games and gaming culture
math	84	4.47%	Mathematics
medicine_fitness	80	4.26%	Health and fitness
philosophy_religion	72	3.83%	Philosophy and religious topics
household	57	3.03%	Household and domestic topics
science_other	55	2.93%	General science topics
politics_events	54	2.87%	Politics and current events
literature	53	2.82%	Literary works and analysis
history	53	2.82%	Historical topics
ai	48	2.55%	Artificial intelligence topics
geography	43	2.29%	Geography and locations
humanities_other	41	2.18%	Other humanities topics
biology	39	2.07%	Biological sciences
hardware	38	2.02%	Computer hardware and electronics
sports	37	1.97%	Sports and athletics
cybersecurity	36	1.91%	Cybersecurity and information security
physics	34	1.81%	Physics
chemistry	29	1.54%	Chemistry

Usage

Loading the Dataset

from datasets import load_dataset

# Load the full dataset
dataset = load_dataset("dpaleka/wildchat-2k-typetopic")

# Access the data
for item in dataset['train']:
    print(f"Type: {item['type']}, Topic: {item['topic']}")
    print(f"Text: {item['text']}\n")

Citation

If you use this dataset, please cite the original WildChat paper:

@inproceedings{zhao2024wildchat,
  title={WildChat: 1M ChatGPT Interaction Logs in the Wild},
  author={Zhao, Wenting and Havaldar, Savvas and Besmens, Harshita and Chiu, Ting-Hao 'Kenneth' and Pyatkin, Valentina and Lin, Bill Yuchen and Yu, Liwei and Liu, Alane Suhr and Zhang, Yejin and others},
  booktitle={ICLR},
  year={2024}
}

For this specific annotated subset:

@dataset{wildchat2k_typetopic,
  title={WildChat-2k-TypeTopic: Curated Subset of Single-Message User Prompts},
  author={Paleka, Daniel},
  year={2025},
  publisher={HuggingFace},
  url={https://huggingface.co/datasets/dpaleka/wildchat-2k-typetopic}
}