[ExI] Singularity news

[Jason Resch]

On Wed, Apr 19, 2023, 4:21 PM Giovanni Santostasi via extropy-chat <
extropy-chat at lists.extropy.org> wrote:

> *I just read a physics article about symmetry... would you be willing to
> assert the point above with additional clarity on "language"?  Either by
> defining it as some form of protocol with agreement on semantics or
> reducing it to simple signals that are (arbitrarily) constrained by time
> and space?*
>
> Hi Mike,
> One of the interesting things is actually the LLMs are not told anything
> about semantics or even grammar but they actually derive it from the
> language regularities. It is not obvious at all and human experts claimed
> it was not possible for neural networks to do that from scratch and direct
> programming. These experts were wrong.
> They simply have no understanding of how the ANN work, they are supposed
> to come close to how our brain works (plus much better memory and
> processing power than we have) and extract meaning from patterns and
> relationships (what else is ANN if not a bunch of knots in a complex
> tapestry?). The meaning emerges from the patterns and no referents (or at
> least absolute, external referents) are necessary.
> In physics, this is true even more so and I gave many examples in the last
> several weeks of this ongoing discussion.
> I wanted to write a Medium article on this. The fact we are using language
> models to understand the world is a powerful theoretical unification too
> and when you get this unification in science a lot of progress follows
> naturally. I have also a strong intuition that actually language gave us
> the consciousness we associate with being humans.
>
> Giovanni
>


To add to Giovanni's points, as I see it, every instance of intelligence is
reducible to some form of prediction. That is, intelligence (selecting an
intelligent course of action) is always done by predicting which of a set
of possible actions will yield the most favorable outcome. I write about
this a bit here:

https://alwaysasking.com/when-will-ai-take-over/#What_is_Intelligence

Framed in this light, LLMs, as general purpose systems able to learn
patterns from examples and create models that can predict future symbols
given past samples, can be viewed as algorithms for both universal learning
and universal intelligence, which is not limited to the domain of language,
but applicable to any kind of behavior that requires intelligence.

Jason





>
> On Wed, Apr 19, 2023 at 8:17 AM Mike Dougherty via extropy-chat <
> extropy-chat at lists.extropy.org> wrote:
>
>> On Wed, Apr 19, 2023, 8:15 AM Giovanni Santostasi via extropy-chat <
>> extropy-chat at lists.extropy.org> wrote:
>>
>>> . Everything is language even for our brains.
>>>
>>
>> I just read a physics article about symmetry... would you be willing to
>> assert the point above with additional clarity on "language"?  Either by
>> defining it as some form of protocol with agreement on semantics or
>> reducing it to simple signals that are (arbitrarily) constrained by time
>> and space?
>>
>> Hmm.
>>
>>> _______________________________________________
>> extropy-chat mailing list
>> extropy-chat at lists.extropy.org
>> http://lists.extropy.org/mailman/listinfo.cgi/extropy-chat
>>
> _______________________________________________
> extropy-chat mailing list
> extropy-chat at lists.extropy.org
> http://lists.extropy.org/mailman/listinfo.cgi/extropy-chat
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.extropy.org/pipermail/extropy-chat/attachments/20230419/ea23dae3/attachment.htm>