Text-to-World Bench
Can a short message become a completed real-world task? We tested text agents on reservations, food delivery, movie tickets, and flights.
Explore the benchmark7 agents11 tasks
Delphi / Research
Research.
Dancy-bench
Can we teach AI a flexible moral framework so it can reason through new situations and avoid harmful outcomes? We worked with postdoctoral researchers in philosophy to test that question.