Move 1: the model writes the converter, it does not read the file
Lesson 2 of 6 · What the model can and cannot see
1 · Learn the move · Specific task + easy yes
Move 1 is the single reframe almost nobody arrives holding: the model is the author of the conversion script, not the reader of the file. Once that lands, the vendor-binary wall stops being a wall - .czi to OME-TIFF, .raw to mzML, proprietary plate exports to csv, all of it is a small script the model can draft in a minute. The technique is a specific task with a checkable yes: not "help me with my imaging data" but "write a script converting this one .czi to OME-TIFF using this named library, with three checks built in - input exists, output is nonempty, and one property I care about survived." A script whose success you can verify is a task the model does well; a vague plea is a task nobody does well. Ask small, check the result, then scale to the folder.
Write a [language] script that converts [one specific file] from [format] to [open format] using [named library]. Requirements: (1) fail loudly if the input is missing or unparseable - no silent skips; (2) after writing, verify the output is nonempty and re-openable; (3) read [one property I care about, e.g. pixel size, channel count] from input and output and print both - if they differ, exit with an error. Print a one-line summary. No other features - I will extend it to a batch after this one file works.
2 · Your turn. You write the prompt
You have one Zeiss .czi confocal stack and a collaborator who needs OME-TIFF by Friday. You have never converted either format, and the microscope facility's wiki page was last updated four years ago. Write a prompt that gets you a checkable single-file converter, not a lecture on bioimage formats.
Remember: the AI sees only your prompt, not this page. If the situation isn't in your prompt, it doesn't exist.
Optional. These shape the output when you run your prompt below, not your score.