On 10/27/26 5:19 AM, Ben Finney wrote:
When a machine is unreliable, we have no obligation to suffer it. We are
right to complain that it is used to impose on our work and our
attention.
But here, it's right most of the times. And you're not the slave of the
machine, it's the opposite: we're giving you a tool. You are right to
complain that it's not perfect, but you are wrong of you declare it
useless, because it simply isn't. LLMs aren't your brains, they are your
3rd hand.
On 10/27/26 5:19 AM, Ben Finney wrote:
>> So if you think it's a waste of time, feel free to ignore the report.
> Not good enough. This is the same logic of the spammer: “I'm just
> sending the message from my automated machine that doesn't care
> whether you want the message; if you feel it's not for you feel free
> to ignore it” does not fly for the spammer, and it does not fly in
> this case either.
A spammer is trying to fool you to get some benefits from you. Here,
you're being helped by the message. You're free to ignore it, but it
really isn't in your interest to do so. Nobody will actually blame you
for not reading it. Well, of course except if you've done some obvious
packaging mistakes, the LLM tells you about it, and you're wasting the
DFSG team time: in such case, you may be blamed, IMO rightly.
That being said, I agree with your opt-in thing. Not because of LLMs
themselves, but because I receive A WAY too many emails from these
automated reviews. I've been receiving again and again mails for
packages I didn't upload, or didn't even care, just because I'm part of
the team. That's really annoying. And I don't see the point of sending
them by email, when they are already available on the web.
On 10/27/26 5:19 AM, Ben Finney wrote:
> Unlike the tools that existed in Debian workflows long before, LLMs
> are acknowledged to be non-deterministic in their output
But their output is never the less useful.
Oh and by the way, lintian, written by humans, produces A LOT of false
positive. To the point it's very common that we write some
lintian-overrides...
> and infeasible to explain exactly how they produced that output.
That, I strongly don't agree. It shows you don't understand how LLM
works. I suggest you read more about them. It's not magic, it still is
computer science, even though it has to do with statistics.
Cheers,
Thomas Goirand (zigo)
