[ExI] Another ChatGPT session on qualia
gsantostasi at gmail.com
Thu Apr 27 06:51:04 UTC 2023
I'm using GPT-4 and when I want to test its spatial relationship I ask
GPT-4 to write code in vector graphics. Then I copy and paste the code in
HTML and I see the results.
They are pretty amazing and this is the sort of the experiments that
actually can really tease the cognitive abilities of these systems. To me
the fact that GPT-4 can go across media, understand where the tail of a
helicopter is, draw a cockpit when you ask to add a human (and it can
explain why it did it), and use superimposed circles to describe the
fuzziness of clouds is completely mind-blowing. Claiming there is no
understanding in these operations is just being in some high-level denial
and cognitive dissonance.
One think you can try is to look for puzzles maybe or some of the typical
theory of mind tests that psychologists perform on young children. Some
scientists have done a very systematic study of these capabilities and
concluded that GPT-4 has the theory of mind of 9 years hold human.
On Wed, Apr 26, 2023 at 11:45 PM Adrian Tymes via extropy-chat <
extropy-chat at lists.extropy.org> wrote:
> On Wed, Apr 26, 2023 at 10:55 PM Giovanni Santostasi <
> gsantostasi at gmail.com> wrote:
>> Adrian try to develop clever tests to determine GPT-4 cognitive
>> abilities. Also I see you use GPT-3 or 3.5 that is vastly different from
>> GPT-4 in terms of capabilities.
> I use what I have access to. My experiments so far have been via
> https://chat.openai.com/?model=text-davinci-002-render . Is there
> somewhere else I should try? Note that I have a budget of $0 for this, so
> anything that requires me to pay is right out.
>> Did you see some of my cognitive experiments? In particular, the one
>> where I asked to draw objects using vector graphics?
> Yes. I suspect you are using some other interface.
> I make no apologies for being more of an experimentalist than a theorist,
> which is why I have been ignoring most of this conversation. This is why
> I was looking for ways to show that experiments with actual AIs could be
> valid at all, rather than dismissed on theoretical grounds. Based on the
> responses, I am not sure how well that worked.
> extropy-chat mailing list
> extropy-chat at lists.extropy.org
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the extropy-chat