[ExI] ChatGPT-4 gets the joke, almost

[Gordon Swobe]

On Tue, Apr 25, 2023 at 4:51 AM Giovanni Santostasi <gsantostasi at gmail.com>
wrote:

>
> *Why do you suppose Sam Altman and Ilya Sutskever, founders of OpenAI,
> believe the only good test of a conscious LLM would involve training it on
> material completely devoid of references to first person experience?*I
> head them saying this and it makes no sense at all. The fact they are the
> founders of OpenAI doesn't mean they really understand deeply how LLM work,
> they may be aware of the technical details of how they are structured and
> how they operate but not necessarily how they achieve their abilities given
> really nobody knows.
>

I see. You believe you understand large language models better than the
founders of OpenAI, the creators of the state of the art language model.
The test was proposed by Ilya Sutskever, one of the top computer scientists
in the field, then endorsed by Altman. Our old friend Eliezer Yudkowski
also endorses it. But I'm sure you know better than these people.

What is implied by the suggested experiment above is that GPT-4 arrived at
> a sense of self...
>

You denied my claim you believe there is a ghost in the machine, but there
it is in your belief that GPT-4 actually has a sense of self and is not
merely writing as if it has one.  You believe there is somebody there.

> You believe yourself superior...

Depends on what you mean. GPT-4 is far more "intelligent" than I am in many
ways, but I do not believe that because it was trained on so much material
written by conscious people and it can do a reasonably decent job of
mimicking consciousness, that it actually is conscious. I do not believe in
ghosts. More exactly, I do not make the mistake of anthropomorphizing what
is only a tool.

-gts
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.extropy.org/pipermail/extropy-chat/attachments/20230425/18b501cd/attachment.htm>