Zuckerberg goes SCORCHED EARTH. Llama 3.1 BREAKS the AGI Industry

[John Clark]

It looks like Mark Zuckerberg is going to do to AGI what Linus Torvalds did to Unix, and I am delighted. It also looks like during the next few years the big money will not be made by software companies but by hardware companies such as ASML, Nvidia and TSML. And lack of powerful hardware is the one disadvantage that China has.

Zuckerberg goes SCORCHED EARTH. Llama 3.1 BREAKS the AGI Industry

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

zma

[PGC]

In Silicon Valley history, there are few spectacles as tantalizing—and as perilous—as the sight of countless billions being funneled into a single technological frontier. AI is, next to impressing us all, also starting to show signs that it might just be the next bubble waiting to burst.

Despite the euphoric predictions and benchmark crunching we hear for awhile now, a growing chorus of investors and analysts are starting to pump the brakes. Goldman Sachs' Jim Covello recently remarked, "Despite its expensive price tag, the technology is nowhere near where it needs to be in order to be useful." Google's latest earnings report underscores this sentiment, revealing very modest profit margins and surging costs. The tech behemoth plans to spend $49 billion on capital this year, an eye-watering 84 percent increase over its historical average.

Google's CEO, Sundar Pichai, insists that the risk of underinvesting is greater than that of overinvesting, a stance echoed by his peers at Microsoft and Meta. Yet, the stark reality remains: AI is burning through cash at a staggering rate, with little revenue to show for it. This year alone, investors are expected to pour $60 billion into AI development—enough to create 12,000 products the size of OpenAI's ChatGPT. Do we need that many? Is this the paradigm now?

The parallels to past tech bubbles are striking, such as dot-com crash and the autonomous driving hype of 2017. Capital continues to be vacuumed into the AI sector with very little to no attention being paid to company fundamentals.

What of the immediate financial hemorrhaging—such as OpenAI's projected $5 billion loss this year?

So, with all this in mind, is it still the right time to invest in a cluster of Nvidia A100s and walk boldly into the future? To the computer and AI specialists reading this: Do you still see the incredible potential in large language models (LLMs) that justify the current hype and investment? Has this technology made your lives easier, or made you 10x more productive, fulfilling Jensen Huang's prophecy that "People don't have to learn programming anymore to develop software themselves"? And finally, where are all the killer apps built by hordes of agents conferring with each other and debugging and refining their code constructively? 

[John Clark]

On Sat, Jul 27, 2024 at 1:29 PM PGC <multipl...@gmail.com> wrote:

" The parallels to past tech bubbles are striking, such as dot-com crash

The two things are not comparable. The invention of the World Wide Web was an interesting development, but AI is more profound than the invention of fire. There is quite simply nothing in human history they can compare with the recent developments in AI and what we will see in the near future, and by that I mean in the next 10 years or less, possibly much less.    

> AI is burning through cash at a staggering rate

Yep. That's the problem, if you're a software company and you spend a lot of money on AI you MIGHT go bankrupt, but if you're a software company and you DON'T spend a lot of money on AI then you're CERTAIN to go bankrupt.

 

> AI is, next to impressing us all, also starting to show signs that it might just be the next bubble waiting to burst.

Thanks to Zuckerberg and the rise of open source, AI software companies like Open AI and Anthropic may go belly up, but not hardware companies like ASML, TSML and Nvidia. Open source or closed, *somebody* (maybe private companies, maybe government. maybe the military) is certain to be developing and running AI programs that require a gargantuan amount of computation, and they will need hardware to do that, so companies that are involved in that should do well. Although TSML might go out of business but not because AI fizzles out but because of a Chinese invasion of Taiwan.

 > This year alone, investors are expected to pour $60 billion into AI development—enough to create 12,000 products the size of OpenAI's ChatGPT.

Most of those products will flop but a few will grow exponentially. 

 > Do we need that many?

We do unless you can figure out a sure fire way to separate the good from the bad.

 > Jim Covello recently remarked [...]

Do you really think Jim Covello knows what's going on? 

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

9$z

z

[John Clark]

That should be TSMC not TSML.

John K Clark

[Tomasz Rola]

On Sat, Jul 27, 2024 at 10:29:29AM -0700, PGC wrote:
> In Silicon Valley history, there are few spectacles as tantalizing—and as
> perilous—as the sight of countless billions being funneled into a single
> technological frontier. AI is, next to impressing us all, also starting to
> show signs that it might just be the next bubble waiting to burst.

[...]

>
> The parallels to past tech bubbles are striking, such as dot-com crash and
> the autonomous driving hype of 2017. Capital continues to be vacuumed into
> the AI sector with very little to no attention being paid to company
> fundamentals.
>
> What of the immediate financial hemorrhaging—such as OpenAI's projected $5
> billion loss this year?
>
> So, with all this in mind, is it still the right time to invest in a
> cluster of Nvidia A100s and walk boldly into the future?

Depends. Do I have something that would require A100 in order to give me
results? Other than Chad Geppetto which is amusement, and not very
useful. I would not dare to have a database on my computer which
invents me non-existing results. If I had to install it, I would not
dare to use it. Chad Geppetto is, to me, software equivalent of
certain kind of people, who always are right, even if they have to lie
to prove it. Good for some entertainment. Not good for building
bridges, aeroplanes and medicines.

> To the computer and AI specialists reading this: Do you still see
> the incredible potential in large language models (LLMs) that
> justify the current hype and investment?

It looked like something big, until I learned about hallucination
without brakes. Until I have read about self driving car speeding up
to drive over a woman crossing the road. Ok, so you can teach it to
not drive over this one particular woman crossing this one particular
place. What about other situations, other people?

Maybe building bigger LLM is the answer or maybe not. There is no
proof, so far, but there is plenty of experimenting on live subjects.

> Has this technology made your lives easier, or made you 10x
> more productive,

I have heard some people tried this. I believe it is akin to asking an
intern to writing me some initial code. After that I still need to
spend time looking for errors in the code and maybe also bending it
another way, because "intern" might not understood. Because Chad is
not equal to the person who sits in your business day and night, ten
years in a row.

It might work in some generic cases - something generic enough to not
require business specifics.

Oh and now they want Chad to remember. My questions would be
remembered. My questions may leak out. Something specific to what I
wanted to do may leak out... days or years later. No thanks. There is
going to be a whole insudrialt egaspione branch related to milking
Chad. People who used Chad's help when planning some future company
move may get sued for the leak or may have to prove they did not cause
the leak, resulting in smaller revenues.

> fulfilling Jensen Huang's prophecy that "People don't have
> to learn programming anymore to develop software themselves"?

A long living phantasy, wet dream of people who will not learn to
program (and will not want to pay to those who want to learn,
maybe?). Look, we all have some specialty. Do I want to be a
neurosurgeon, with Chad Geppetto doing all dirty/bloody work in my
name and me getting the hard earned cash?

> And finally, where are all the killer apps built by hordes of agents
> conferring with each other and debugging and refining their code
> constructively?

Good question.

--
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 **