LLM policy

[Sandro Santilli]

On Sat, Jun 13, 2026 at 07:14:25PM -0400, Greg Troxel wrote:
> 
> Many people have concluded that all extant LLMs are bad.

Sorry, I didn't get this statement.
Who are "many people" ? I've heard by many people that LLMs are good too.

Personally I find good uses and even some fun with LLMs.
Locally downloaded, legally obtained, used by free software inference tools.

Also, although I'm not an expert in this field, I understand there are
many different LLMs, some of which specifically built (trained) to encode
GIS specific knowledge.

What's fascinating to me about LLMs is that we (humanity) found a way to
encode so much knowledge (debatable term) into so few bytes.
I have experienced a < 4GB being able to generate coherent (and in some
cases even correct) answers to questions I've asked in different languages.

Yes, they are just a blob of statistics and all we get out of them are
probabilistic outputs, so what ? Why does this make them "bad" ?

Besides, yesterday a human friend told me:

  "I know what you are going to answer"

does that makes me a stochastic parrot too ?

> if we had people who were interacting
> wtih postgis in order to gain trust and plant a vulnerability (xz
> style), then we'd say that wasn't ok, and we would almost certainly not
> be having a debate.

Agreed.

Are we having a debate about such a specific behavior you saw in PostGIS ?
Or are you suggesting that "allowing use of LLMs" undermines the trust you
have in the PostGIS source code ?

> But that isn't the question on the table for projects.  It's whether
> code produced by LLMs is acceptable

I'd say it depends on the code ?
Is code produced by any contributor acceptable ?
We (the maintainers, those who have write access to the official repo)
decide whether or not to accept it,
and we take the responsibility of doing so.

> There's a further question, which is if full-on open-loop agent
> interaction is acceptable.

I think what people do on their free time is not our business ;)

> (That's before one considers the ethical issues of using LLMs trained on
> stolen code, for many, the environment costs of LLMs, and for some, the
> coming financial fraud of IPO to index conversion before the bubble
> bursts as companies that can't possibly make money fail.  I consider it
> unethical to use LLMs that are trained on stolen data or data from
> aggressive scraping.  That's a personal judgement, and I realize
> opinions differ.)

I do share your concerns here, but I need to know more before I can really
have an informed opionion. So far all I experienced was the the environment
cost (it is very often a lot faster and lighter to directly do things rather
than asking an agent to do it for you).

--strk; 

  Libre GIS consultant/developer 🎺
  https://strk.kbt.io/services.html
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 659 bytes
Desc: not available
URL: <http://lists.osgeo.org/pipermail/postgis-devel/attachments/20260618/18bf225a/attachment.sig>