Reminder: #LibreOffice includes no AI features in the software. It doesn't rewrite your content, train algorithms, or share your data. There is zero AI functionality in the app.
Regarding the use of LLMs in the development process, we have a strict policy about this: wiki.documentfoundation.org/Development/AI_policy – Contributors who use such tools must be able to explain EVERY changed line of code. We do not accept any "slop".
@libreoffice So why don't you enforce your "strict policy" on submissions? I went through some random LibreOffice core commits and found three that used LLMs. None of them actually followed this policy.
@barubary@libreoffice I'd be interested in specific examples, can you share those and the exact violations? Please don't get me wrong here, I'm not pro-LLM or want to defend LO here, I just want to see that myself.
@serak@libreoffice "You must be able to ..." is generally unverifiable. The hard requirements in this policy as I see them are: The commit message must mention the name of the LLM product. The commit message must mention the version of the LLM product. The commit message must document "the steps taken", which I'm not sure how much detail they want, but I'd expect at least a high-level description like "I wrote prompt/specification X and told product Y to generate code" or "I originally wrote this code by hand, but I had product X review it and it pointed out the following issues, so I changed it" or something like that. The commit message must list the line numbers where an LLM was "involved", which I'd say includes all LLM-generated code and probably also anything written/changed after LLM review.
From what I can see, none of the LLM commits do (3) or (4). (Disclaimer: I'm not sure how useful a list of line numbers would actually be in practice, but the policy says you must have them.)
