On Tue, Oct 27, 2026 at 04:39:44AM +0100, gregor herrmann wrote: > |This is an LLM based review using a Debian Inference based workflow. It was > |not written by a member of the DFSG team and has not been checked by one. It > |is not a decision on the package: it may be wrong, and it may have missed > |things. > > > I do not object in principle to the use of a LLM, but the DFSG team > > needs to take full responsibility for its output: this kind of > > disclaimer is not acceptable. > > Same here: I don't object to the DFSG team experimenting with LLMs for > package reviews but dumping them without any human inspections onto > uploaders seems both disrespectful and contrary to the GR result to me.
+1. In general, my approach to LLM usage is that its output should be directly consumed only by the direct user of the LLM. Any downstream consumption should come only after it's been filtered through a human. It's well documented that LLMs can make silly mistakes. It is the responsibility of the LLM user to screen for such mistakes, not the person with whom the user is collaborating. So in the case of NEW processing, IMHO the best usage of an LLM would be to highlight aspects of the package to the DFSG team for closer scrutiny, *not* to generate email directly to the maintainer. This still makes this potentially powerful tool available to an overworked team while also ensuring that package reviews are of high quality. noah
