DatumFormsChoose the better response.
Compare model outputs side by side. Learn which answer people find more helpful, relevant or easy to understand.
Pairwise preferences
FOR AI TRAINING & EVALUATION TEAMS
Better AI starts
with human insight.
Turn human preferences, ratings and explanations into data for your next model iteration. Reach a diverse participant panel without recruiting and managing your own pool.
Human feedback · An existing participant panel · Your own workflow
FEEDBACK YOU CAN PUT TO WORK
Find out what people
prefer. And why.
Collect the judgments your team needs to train, compare and evaluate AI systems.
Measure how well it works.
Collect consistent ratings for clarity, usefulness, tone and other criteria that matter to your product.
Human evaluation scores
Give your data more context.
Gather categories, sentiment and written explanations to support your training and evaluation datasets.
Labels and annotations
Spot what your model misses.
Hear how people interpret an answer in the context of their own experience. Uncover assumptions a score alone might miss.
Qualitative insights
THE PEOPLE ARE ALREADY HERE
More perspectives.
Less recruitment work.
Spend your time improving your model. Use our existing panel to reach people with different backgrounds and lived experiences.
Find the audience for your task ↗Skip building a participant pool.
Bring your questions to an existing panel, without sourcing a fresh recruitment list for every round.
Hear from different people.
Collect feedback across age groups, regions and demographic backgrounds to understand a wider range of experiences.
Make your budget go further.
Affordable participant access helps you gather more feedback across the tasks and model comparisons that matter to your team.
FROM TASK TO DATASET
Human feedback.
In your workflow.
Connect survey creation and response retrieval to your existing tools through the API.
- 01
Bring your evaluation tasks.
Create questions around your prompts, candidate outputs and evaluation criteria.
- 02
Reach the right people.
Match your task to the audience and response volume you need through our participant panel.
- 03
Collect structured feedback.
Pull preferences, ratings, labels and explanations directly into your data pipeline.
- 04
Keep improving.
Compare model iterations and run new feedback rounds as your questions and evaluation needs evolve.
BUILT TO CONNECT
Bring the feedback to your stack.
Create surveys programmatically and retrieve responses through one API. Keep your datasets flowing into the tools your team already uses.
HUMAN INSIGHT FOR YOUR NEXT MODEL
What should your AI
understand better?
Bring your tasks. Reach our panel. Put human feedback to work.