How it works
You type a maths question. An AI writes the answer as numbered steps, and finishes by saying which parts of its own argument it was unsure about. The site shows you that list first, before the working, so you know where to look hardest.
Why that matters
AI is good at sounding certain. A wrong step reads exactly like a right one, which is why people either trust these answers too much or not at all. Making the AI point at its own weak spots does not make it correct, but it tells you where to start checking — and that is usually one or two steps, not the whole page.
What the marks mean
The two buttons on each step
Why is this true? asks the AI to name the rule or theorem behind that one step. Challenge it asks the AI to attack its own step and look for a case where it fails. Both are useful, and both have the same limit: you are asking the same AI that wrote the step. It catches carelessness. It does not catch things it never knew.
Things worth knowing
- If nobody has used the site for a while, the first answer takes up to a minute while the server wakes up.
- There is a daily limit on questions. This is a free trial running on a small budget.
- Very long answers can get cut off. When that happens the page says so at the top rather than pretending the answer is finished.
- Your question is sent to an AI company to be answered, and we keep a copy so we can see which questions the site handles badly. Do not type anything private.
- Your questions and answers are also saved in your own browser so they are there when you return. The "Delete saved questions" button removes them.
What comes next
Real checking by computer: putting numbers into the formulas the AI claims, testing small cases, and hunting for counterexamples. That is what turns a grey "AI-written" step into a green "checked" one, and it is the next thing being built.