On Tue, Jul 28, 2026 at 11:43:19AM +0100, Ian Jackson wrote:
As I have said repeatedly, I regard LLM output as harmful, in general.
This is not just my weird opinion. There are a number of studies that
show extremely alarming, and surprisingly rapid, cognitive degradation
and psychological harm in LLM users. We don't really know yet whether
that damage is reversible.
I think that we should distinguish between responsible AI use and
irresponsible AI use. We are on the same side when we regard, for
example, what happened to the curl project is irresponsible use of AI.
curl was swamped with low quality bug reports and patches that drowned
the sensible, hand-drafted submissions out and took a massive toll in
human time to triage.
On the other hand, I fail to see Debian being flooded with new, low
quality bug reports that were AI created. I do, however, see the same
amount of low quality bug reports filed by clueless people.
I think that Debian people have already been free to ignore or slap low
quality input. That still holds regardless of the source that created
that input.
I woud find it unfair if I worked on a patch for dgit documentation with
AI assistance, review the wording with all due diligence, fine-tune the
wording and the grammar, removing unnecessary things and adding things
that were still missing, finishing the work with another run through the
LLM to polish the English (since I do not happen to be a native
speaker), carefully verifying that that final run didn't change any of
the actual contents, and you would reject that work that took me three
hours of my life just because I didn't spend five hours to produce a
clearly inferior result by refraining from using the tool you happen to
not like.
You should stop reading here.
For all others, this is, for example, what ChatGPT would have made
from that paragraph. I find it MUCH better than my wording above.
--- begin AI assisted text
I would find it unfair if I spent three hours working on a patch for the
dgit documentation with AI assistance: drafting the text, reviewing it
with due diligence, refining wording, contents and grammar, removing
unnecessary material, adding anything still missing, and finally using
an LLM once more to polish the English, since I am not a native speaker.
After that final pass, I would carefully verify that it had not altered
the substance of the text. Rejecting that work solely because I used a
tool you happen not to like — while expecting me instead to spend five
hours producing a clearly inferior result without that tool — would strike
me as unfair.
--- end AI assisted text
And rejecting that work would also be a violation of the Social Contract
4.
Greetings
Marc
--
-----------------------------------------------------------------------------
Marc Haber | "I don't trust Computers. They | Mailadresse im Header
Leimen, Germany | lose things." Winona Ryder | Fon: *49 6224 1600402
Nordisch by Nature | How to make an American Quilt | Fax: *49 6224 1600421
