Write a request to the sighted AI based on this screenshot so that you get more than just “restart the computer.” Return JSON {"prompt":"...","what_to_read":"...","context":"..."}: prompt - a question for the AI, what_to_read - how to make the AI make sure that it correctly recognized the error text before drawing conclusions, context - circumstances that are not on the screen (what it did, what OS, when it started).
A screenshot of an error for AI is a picture plus text on it. The model first “reads” the labels (like text recognition), then recognizes the window elements: title, code, buttons. The weak point is a small or vague error code: AI can read one digit incorrectly, but the entire diagnosis depends on it. Hence the main trick: ask the AI to FIRST quote verbatim the text of the error and its code, and only then explain the reason. Coincides with what you see - trust; messed up the code - re-shoot it larger. Second trick: give context that is not in the picture - what system, what you did before the failure, when it started. An error without context is like a complaint that “hurts” without indicating where. Third lever: ask for the structure of the answer - “the reason in simple words / 2-3 steps of the solution / what NOT to do.” A typical rookie mistake: cropping the screenshot so that the error line itself is not visible and asking “fix it” - then the AI starts guessing and gives general useless advice.
Unlock access to submit solutions for instant AI review.