No subject

[Regina Obe]

> >> > LLM Policy - how do we answer to people why we don't need an LLM
> >> > Policy
> >>
> >> Can you clarify if in the minutes there was a preceding missing:
> >>
> >>   the group decided that there is no need for an LLM policy
> >>
> >>   the group agreed that existing rules imply [something] (specifically
> >>   about if there is or isn't a duty to disclose, and about licensing of
> >>   LLM output)
> >>
> >> Given the list discussion, I am surprised by the one line in the
minutes.
> >
> > That was alluding to a grander scheme of documentation we are working
> > on that provides guidelines on how to develop for PostGIS work whether
> > you are using an LLM or not.
> 
> So there was no discussion of whether or not there should be any policy
> surrunding LLM usage?  And not of whether or not current guidance bears on
> this?  I thought that was on the agenda.
> 

We were mostly focused on packaging issues and how it impacts users and
packagers.
That was what most of our time was spent on.
There was some discussion about LLM and how the barrier between LLM and
human is going to get fuzzier (that was the pre-meeting) as time goes on
just as we can't really imagine doing some tasks without a computer.

> > Darafei is still flushing that out with his experiments as you can see
> > here
> > - https://github.com/postgis/postgis/pulls
> 
> Are those not to be merged, until there is a policy decision?   Seems
> like a headlong rush.

They'll be merged as he scrutinizes them, he isn't just going to pull them
in without serious scrutiny.

The idea is to create scenarios that forces us to think about what we want
in a policy using real world scenarios.

Part of the exercise is 

Imagine you as a user, have a bug, so you want to prompt engineer your way
out of it
So of course you post a prompt engineered pull request.

How does this impact us as maintainers?

Can we set up guidance, that simplifies the work we need to do to minimize
on getting crappy pull requests.  Use AI to solve AI issues.

Can we ourselves just throw our bug tickets at AI and coach it to give us
decent results.

So in the end, we have an idea of what our policy will look like.  Not LLM
specific, more like what a proper pull request should look like.

And hopefully we close out a lot of our bugs in the process.

WIN WIN.