Make quality measurable.
Compare creative work using reward models trained to predict which work people prefer.
Better creative AI
starts with human taste.
Reward models to guide training. Benchmarks to help you find your creative AI.
View the benchmarkLatest research
Benchmark / 01Creative Writing Bench V1
How does AI writing compare with writing by people? Explore models ranked by predicted human preference.
Explore the resultsWhy we do this
Creative work has
no answer key.
AI can follow every instruction and still produce something forgettable. We’re interested in work that holds your attention, feels original, and stays with you.
We study how people judge creative work and turn those preferences into feedback that helps AI improve. Our first tools focus on creative writing.
Our approach
From human preference
to better creative work.
Learn what resonates.
Study what readers value and where model outputs fall short, with human writing as a reference.
Turn insight into progress.
Develop reward signals that can help train AI to produce stronger creative work.
Starting with writing. Building toward other creative tasks.
Start a conversation
Good work starts
with a question.
Interested in creative AI, evaluation, or reward models? We’d like to hear what you’re working on.
Your message is delivered privately and used only to respond to your inquiry.