Policies
Review by hand
Every edit in your review must be read, judged and made by you. Tools can help you check your work, but you verify everything. This page gives the rule, the reason for it, and what to do if AI wrote part of your review by mistake.
This rule replaces any earlier advice you may have seen about AI tools.
Why we ask this
Your review captures what AI does not always get right: cultural nuance, and what is correct for readers in your language and region. That works only when every edit is read and judged by you.
Your corrections are the human reference that we use to grade the translation engine. They show where the engine is right and where it is wrong. Each merged lecture also becomes a reference version that later versions of the engine are checked against.
If AI wrote a correction, we would be grading AI with AI. A review can be the human reference only when the judgement in it is your own. See How your review is used.
Tools that help you check
Dictionaries, spelling and grammar checkers, and references such as Wikipedia, the official Python documentation and textbooks are useful, and you are welcome to use them. Asking colleagues is fine too. Read and verify every change a tool suggests before it goes into your review: a tool may suggest, but you decide.
AI assistants and machine translation are different, because they write the text for you. Do not use them to write or translate any part of your review. That includes AI built into other tools, such as Copilot on GitHub, and the translate feature of web browsers.
If you are unsure about a tool, ask @mmcky.
Reports on the English
One exception: you may use AI tools to help write an issue that reports a mistake in the English lecture, for example to check your English. Such a report goes to the English repository, not into your review. See Errors in the English.
If AI wrote part of your review by mistake
Say so in a comment on the pull request. We then exclude that review when we grade the engine. A short note is enough, for example: “I used an AI tool for part of this review.”
Our side
The drafts you review are machine translations. Our translation engine, action-translation, makes them with an AI model. Each pull request says how its draft was made. See Reading a review pull request.
A pull request may also carry automated reviews, such as the one from GitHub Copilot. These are AI as well. They need no attention from you, and your review should not rely on them. If you agree with something an automated review says, write your own comment in your own words.