[ExI] Why “Everyone Dies” Gets AGI All Wrong by Ben Goertzel

[John Clark]

On Sun, Oct 5, 2025 at 3:24 PM Ben Zaiboc via extropy-chat <
extropy-chat at lists.extropy.org> wrote:

*> Hmm. We're not just talking about software, though. This is very
> complex hardware as well.*


*That's true but even if we don't take nanotechnology into consideration
(and we really should) the dividing line between software and hardware is
starting to become a little blurred. If you divide any modern AI into
smaller and smaller components you'll find that its fundamental operation
is simply matrix multiplication, the sort of thing you probably learned in
high school.   *

*An AI called AlphaTensor found an algorithm for multiplying two matrices
together that was faster than anything humans had ever found. With this new
algorithm computers can multiply two 10 by 10 matrices together 20% faster
than they could with the fastest human derived algorithm that was used
before. Almost as soon as it was discovered chip
companies like Nvidia started devoting dedicated circuitry to perform this
new algorithm. So is this an advancement in software or hardware?*

*Obviously this is a positive feedback loop because in order to operate all
AI's need to do a LOT of matrix multiplication, and with more efficient
ways to perform it you can do more of it so you get smarter AIs that can
find even better algorithms. The following article is from the October 5,
2022 issue of the journal Nature:*

*Discovering faster matrix multiplication algorithms with reinforcement
learning* <https://www.nature.com/articles/s41586-022-05172-4.pdf>

*John K Clark*

>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.extropy.org/pipermail/extropy-chat/attachments/20251006/0f668616/attachment.htm>