[ExI] How AI understands the world

[Adrian Tymes]

On Sat, Feb 21, 2026 at 7:51 PM <spike at rainier66.com> wrote:
> -----Original Message-----
> From: extropy-chat <extropy-chat-bounces at lists.extropy.org> On Behalf Of Adrian Tymes via extropy-chat
> Subject: Re: [ExI] How AI understands the world
>
> On Sat, Feb 21, 2026 at 5:00 PM <spike at rainier66.com> wrote:
> > From: extropy-chat <extropy-chat-bounces at lists.extropy.org> On Behalf
> > Of Adrian Tymes via extropy-chat
> > > ChatGPT was able to view this image just fine.  2 times out of 3 trials I ran, it got the correct answer: B.
> >
> > Cool, what was the answer the third time?  Were there any differences in the prompt?
>
> > C (it got hung up on the arrows, and thought only a left-pointing arrow could be correct).  The prompt each time was, "Can you solve the puzzle in this image?".
>
> Cool, but crazy: identical inputs thrice, different output in one of the trials.  That's as humanlike as anything: my bride is that way.

I only had the patience and energy for 3 tests this afternoon.  (This
past week has been intense and a bit rough for me.)  It would be
interesting to see the results of a larger series of trials, perhaps
across multiple AIs.  If you use one with memory (as I did), make sure
to wipe that memory between each trial (as I did).