Re: Toy/demo: using ChatGPT to summarize lengthy LKML threads (b4 integration)

[Jonathan Corbet <corbet@lwn.net>]

Hannes Reinecke <hare@suse.de> writes:

> Out of personal interest I really have to chime in here.
> You are aware, though, that following this line of thought
> and improving the underlying model (which really shouldn't
> be too hard given the vast body of training data which is LKML)
> we will actually putting LWN and its editors out of business.
>
> I would be _VERY_ interesting having a discussion here with the
> relevant parties (Hello Jon!) to see which direction we as a
> community will want to go. Maybe Plumbers and/or Kernel Summit in Vienna 
> would be a good place to start.

Trust me, if you're in a business that is in any way associated with
publishing, you're aware of such things.

Whether the kernel community goes in for a tool like this will
ultimately have little effect on LWN's fate - others will certainly pick
this stuff up regardless.  There are, after all, certain other sites out
there reporting on the kernel that have looked like bad LLM output for
years anyway...:)

I worry about some of this stuff in general.  It seems there's a
definite risk of creating a checkpatchGPT that is allowed to push
developers around far more than would be warranted, for example.  But
the tools are out there and getting cheaper; if they are useful I guess
we should use them.

It would be nicer to base them on an open-source (or something close to
it) model, of course.

Meanwhile, in the short term I suspect LWN can find a place for itself.
After that I hope to be retired.

Thanks,

jon