Explore WildSeek
Browse the 3,077 annotated queries by risk-sensitive domain, query type and source corpus. See the dataset description below for how each label is defined.
The WildSeek dataset
These are 3,077 information-seeking queries sampled from four corpora of real human–LLM conversations. Annotators labelled each one by hand with the risk-sensitive domain it touches and whether it asks for a fact or for analysis.
Queries per risk-sensitive domain
queries in total
User intent
Is the turn information-seeking at all? There are five classes: information seeking, content creation, coding, no request, not English.
Browse intent labels →Risk-sensitive domain
Domains where a wrong or incomplete answer could affect someone's life, safety, health, money or decisions.
Domain definitions
Open-endedness
Factoid queries can be answered from a single verifiable source. Analytical queries need reasoning, comparison, instructions, prediction or judgment.
Decision rules
- Factoid
- Definitions, historical or scientific facts, descriptions of entities or systems, text lookup. Typical signals: What is / Who is / When did.
- Analytical
- Procedural steps, comparisons and trade-offs, predictions, advice or value judgments. Typical signals: How to / Why / Should I / Best / Compare.
- Borderline
- If one verifiable fact or a concise description answers it, it is factoid. Instructions, predictions and opinions are always analytical.
Explore the annotated queries
Click a cell, row, column or bar to filter. Every panel updates to match the current selection.