DeepMind AI Discovers Better Algorithms for Foundational Computing

[John Clark]

If this is not an example of an AI helping to develop a smarter AI then I don't know what is. People are gonna be having a very hard time trying to convince themselves this is just a glorified autocomplete program. 

AlphaDev - DeepMind AI Discovers Better Algorithms for Foundational Computing

John K Clark    See what's on my new list at  Extropolis

pui

[spudb...@aol.com]

No objection to this at all, as long as it can make new medical discoveries that lead to new cures, innovations in energy that we humans missed, and the lot. If AI cannot help our species, I say its worse than dangerous, it's useless to us!

--
You received this message because you are subscribed to the Google Groups "Everything List" group.
To unsubscribe from this group and stop receiving emails from it, send an email to everything-li...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/everything-list/CAJPayv3Tn98n4yZL61z0Ry__BNvf9WJ25VZOXrrtLAn_fYt0FQ%40mail.gmail.com.

[Brent Meeker]

The next level will be when an AI makes itself smarter.

Brent

[Tomasz Rola]

On Sun, Jun 11, 2023 at 02:52:01PM -0700, Brent Meeker wrote:
> The next level will be when an AI makes /*itself*/ smarter.
>
> Brent

But, but, how are humans going to measure its smartness (to know, for
one example, that it is not faking its own progress).

--
Regards,
Tomasz Rola

--
** A C programmer asked whether computer had Buddha's nature. **
** As the answer, master did "rm -rif" on the programmer's home **
** directory. And then the C programmer became enlightened... **
** **
** Tomasz Rola mailto:tomas...@bigfoot.com **

[Russell Standish]

I haven't read the original article, but given the timing, I believe
this was the application of AI in a competitive environment to
optimise sorting algorithms of short sequences (<8 items, IIUC).

Which reminds of work done by Danny Hillis (of Thinking Machines fame)
in the late 1980s who set up a coevolutionary genetic algorithm to
evolve sorting algorithms that beat classic sorting algorithms. ISTM -
they've just rediscovered his work.

Danny has a paper in the ALifeII proceedings describing that work.

On Sun, Jun 11, 2023 at 02:52:01PM -0700, Brent Meeker wrote:

> everything-list/46449833-251f-cdee-5129-84a7e83f17c5%40gmail.com.

--

----------------------------------------------------------------------------
Dr Russell Standish Phone 0425 253119 (mobile)
Principal, High Performance Coders hpc...@hpcoders.com.au
http://www.hpcoders.com.au
----------------------------------------------------------------------------

[John Clark]

On Sun, Jun 11, 2023 at 7:14 PM Tomasz Rola <rto...@ceti.pl> wrote:

> But, but, how are humans going to measure its smartness 

By using a stopwatch. An Artificial Intelligence has recently found a machine language sorting program that is smaller and faster than anything any human being has seen before. And a Artificial Intelligence has also found a fundamentally faster way to multiply matrices, and nearly every large program needs to multiply matrices at some point, especially if the program involves graphics or, you guessed it, Artificial Intelligence. To me that certainly sounds like the sign post of an exponential process in the making, and exponential processes tend to end in a singularity.

AI Reveals New Possibilities in Matrix Multiplication

to know, for one example, that it is not faking its own progress

In this case it would take more brain power to successfully fake its progress then it would take to actually make real progress.  

John K Clark    See what's on my new list at  Extropolis

mnp

[spudb...@aol.com]

Moreover, as an engineer, how would you apply this and to accomplish what? 

--
You received this message because you are subscribed to the Google Groups "Everything List" group.
To unsubscribe from this group and stop receiving emails from it, send an email to everything-li...@googlegroups.com.
To view this discussion on the web visit

https://groups.google.com/d/msgid/everything-list/CAJPayv36_Ubx0-O5WpkjuCEA5X3wZA07_drmAnX-D3b%2BEOqO8A%40mail.gmail.com

.

[spudb...@aol.com]

Isn't it fair to state that we will measure its intelligence, by what it actually accomplishes? Things that impact us humans.

--
You received this message because you are subscribed to the Google Groups "Everything List" group.
To unsubscribe from this group and stop receiving emails from it, send an email to everything-li...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/everything-list/ZIZVUtS0s2WtftUa%40tau1.ceti.pl.