[ExI] Bender's Octopus (re: LLMs like ChatGPT)

[Rafal Smigrodzki]

On Thu, Mar 23, 2023 at 5:13 PM Will Steinberg via extropy-chat <
extropy-chat at lists.extropy.org> wrote:

> This argument makes no sense though.  Of course the octopus doesn't have
> access to all the information in A and B's brains.  Why would it know
> about bears?  Why would it know how to defend oneself?  Does a baby know
> these things before it has learned them?  Does that make the baby
> non-conscious?  Terrible argument, doesn't show that the AI is not
> conscious or human-like, only that it has less developed sapience than the
> humans, which makes sense, because it has had access to a small fraction of
> the information the humans have.  You might say that it is not conscious
> because it can put together human-looking phrases without having the
> referents you speak of, but what's to say it needs them?  Maybe it took a
> shortcut to meaning by interpolating those referents.
>

### What if that octopus also tapped into the undersea internet cables, and
drank yottabytes of data over the past few years? In order to predict
patterns in data it would have to create representations
of patterns in the data, including a lot of bear pictures and general
knowledge of animals and physical objects. This octopus would know all that
the internet knows about bears and would have a
very good idea about what A and B are talking about.

Rafal
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.extropy.org/pipermail/extropy-chat/attachments/20230324/9cb14115/attachment.htm>