Asking AI for help
The three things Revquix AI will do on a problem, what it costs before you press it, and the four things it deliberately will not do.
Revquix AI appears at three moments on a problem page: a run that errored, a submission that came back wrong, and a solution that has already been accepted. Each is one button and one answer.
The three actions#
| When it appears | What it does |
|---|---|
| Explain this error - after a run or submission that errored or would not compile | Reads the error and your code, and explains what the message means. |
| Explain my failing case - after a wrong answer | Given the failing input, the expected answer and what you returned, explains what your code did wrong on that input. |
| Optimise - only on an accepted solution | Suggests a faster or cleaner version of a solution that already works. |
The middle one is the one worth knowing about. It is specific to the judge - it can see the case you failed on, so it explains a concrete wrong answer rather than reviewing your code in the abstract.
What it costs, before you press it#
The price is on the button. It reads "free ยท 2 left today" or "1 credit" before you click, because a charge somebody did not see coming is a support ticket however fair it was.
- Explaining an error and explaining a failing case are free a few times a day, then cost credits. The button says how many free ones you have left.
- Optimise always costs credits, because it is the expensive one.
- Asking the same question twice is free. If you have already asked this, you get your earlier answer back at no charge, marked as such.
- If you cannot afford it, the button says so and names your balance rather than failing after the click.
The exact numbers live in the product, not in this page. See credits for where your balance comes from.
What it will not do#
This is the part worth reading, because it is deliberate rather than a limitation we intend to lift.
-
It never sees the problem's reference solution. Not redacted from the answer - never given to the model in the first place.
-
It never sees the hidden test cases. The single exception is the one failing case you have already been shown, and only for "Explain my failing case". On a problem where even that case is withheld, the action refuses rather than sending it.
-
Optimise only works on a solution that is already accepted. You may improve something you got working. You may not have it written for you.
-
There is no "fix my code" here. That action exists in the Projects editor and produces a patch you can apply. On a problem, a patch that makes the tests pass is the answer, so it is deliberately absent.
Treat what it says about speed as an opinion#
Every answer carries this line, and it is not boilerplate:
An opinion from a model, not a measurement.
A model asked "what is the complexity of this?" returns a confident Big-O every time, including for code where the honest answer depends on the input. If it tells you something is quadratic, check it against the growth analysis on your accepted submission, which measures rather than guesses.
Related#
Was this page helpful?