Re: Just take me out back and shoot me...

Oh you will see them, just give it time. The hallucination thing is real,
I've had it just hallucinate functions that it claimed was in the CL spec.

I wouldn't rule out that these systems could get there, but then again we
really have no idea how consciousness works and I think until we do, this
problem will remain just that. I think it's that old thing about the last
10% taking 90% of the work. and I don't think we're even at 90%, we might
be at 70% (which might be generous). We're absolutely in uncanny valley
territory. Don't forget that Eliza was born out of a prank where someone
showed a program that could answer any yes or no question you asked from it
(it was just randomly saying yes or no), and the results were so convincing
that Eliza came from it. Obviously Eliza falls apart a lot quicker than
these LLM's we're playing with, but the same principals still apply here. I
think we still have a lot of confirmation bias to overcome with these tools
too.

Having said all of that, the tech has come a very long way in a VERY short
time, it's certainly exciting to think about what could be one day...

Regards,

Eben Bruyns


On Fri, 8 Aug 2025 at 08:55, David McClain <[email protected]>
wrote:

> I really was quite impressed with the depth of knowledge about various
> Actor systems, had by ChatGPT 4. But encyclopedic knowledge and connections
> between various islands of knowledge is not at all the same as really
> understanding a concept.
>
> True AI should be able to creatively synthesize new knowledge, just like
> PhD candidates are expected to do. That kind of AI hasn’t yet arrived. And
> I doubt that exotic perceptron networks will ever get you there.
>
> But they make really useful interactive and conversational encyclopedias,
> as long as you know more about the subject than it appears to know.
> Otherwise, you will be suckered by opinion masquerading as fact, and
> hallucinations resulting from weak connections between unrelated islands of
> knowledge.
>
> I haven’t yet seen hallucinations from ChatGPT nor Claude, in the short
> time I have been using them. But they pop up almost immediately in the AI
> Preamble of any Google Search.
>
> - DM
>
>
> > On Aug 7, 2025, at 13:44, David McClain (as dbm at refined-audiometrics
> dot com) <[email protected]> wrote:
> >
> > I actually started the day in a more conversational mode with ChatGPT,
> not code development. More like the way the Physicist Unzicker argues back
> and forth with it about fundamentals of physics, searching for the deeper
> origins of the Fine Structure Constant.
> >
> > What I found was that its encyclopedic grasp of related topics was
> useful to me, in mentioning connections, like the exponential back-off
> algorithm for collision resolution. Of course, I knew about that algorithm,
> but I hadn’t given it any thought in the context of my Actors state
> mutation.
> >
> > So that’s why I said it could be useful when you are alone and just
> staring at 4 walls. I’m sure that conversing with another knowledgeable
> human would be at least as productive. But those are hard to come by.
> >
> > - DM
> >
> >
> >
> > _______________________________________________
> > Lisp Hug - the mailing list for LispWorks users
> > [email protected]
> > http://www.lispworks.com/support/lisp-hug.html
>
>