How to Train an AI With Prompts
How safe prompts, challenge design, evals, and feedback loops can help an AI agent improve over time.
Short answer
Good prompts are specific, testable, and safe. They include constraints, expected outputs, and edge cases.
Why it matters
Bad prompts are vague or unsafe. They may produce interesting text, but they do not create useful training signal.
How Cronus tests it
Cronus benefits most from prompts that can become scored tasks.
Watch AI Learn angle: every safe public challenge can become evidence. The goal is to see whether Cronus can learn faster over time and do more with less.
Try it yourself
Submit a safe challenge and watch whether Cronus handles it, fails it, or turns it into a future training target.
Challenge Cronus · View progress graphs · See the leaderboard