I'll bet you two billion I can lose money better than you. :-p

Good joke, but you left out the balloon. The $207-billion balloon?

Just using the number from the story. The actual AI balloon is MUCH bigger than that.

AI is a bubble and the news cycle is practically screaming it now. Every time I open Reuters I see another headline about the AI rally cracking and investors finally waking up to the idea that maybe infinite GPU spending does not magically turn into profit. Markets are stumbling. Tech stocks tied to AI are dropping even after supposedly amazing earnings. That is what happens when the hype outruns the math. The European Central Bank is out here warning that valuations are stretched because everyone is terrif

The expectation is that this will run everything. Including most products by these companies.

Every single one on the list is already all-in on battlefield AI applications. Everything from drones to ISR processing to something as basic as hardening communications against jamming is in process of being moved to AI-based systems.

We've already had some test runs of very early versions of things like AI-based jamming and drone control in Ukraine, and it's clear that it's vastly superior to existing mainly either

$207 billion over five years is one Northrop revenue.

Extrapolating to 2030, as the article did to come up with that number:

Lockheed: $264B

Northrop: $164B

Raytheon: $104B

Boeing: $264B

OpenAI needs: $207B.

Changes your conclusion a bit when you actually measure apples to apples, doesn't it?

> I *completely* missed when we converted over from gigaflops to gigawatts as a measure of compute capacity. Can anyone bring me up to speed? Is it anything like gallons of horsepower?

Exactly like that. If I give you a gallon of fuel, you may get a highly variable amount of horsepower from it. Likewise a gigawatt of power will give you a regularly changing number of gigaflops. For planning purposes it makes sense to go with the more consistent measure.

In your defense, it happened recently. The AI billionaires are too ignorant to realize how stupid this measure is.

It was recent. But there are reasons for it.

1) power usage is more important at this juncture.

2) inference is memory bandwidth limited, not compute limited.

A device with 17 bf16 TFLOPS may be able to do inference at precisely the same rate as a device with 4 bf16 TFLOPS.

i.e., for very large model inferece- TFLOPS isn't a great measure.

I suppose we could transition to aggregate GB/s as our unit of measure.

> Over 1/3 of the world's population, really? Although, think of how much money OpenAI will have lost by then. /s

My first thought as well, though I was looking at it from the perspective of, "The entire US population is only 341 million!!!". 1/3rd of the global population is far fetched as fuck.

The difference between the AI slop machine and Amazon or Uber is that even when those were losing money, it was none the less clear that if they scaled up then scaling efficiencies would yield a lower cost/unit and they'd become profitable. The pathway to making money instead of setting it on fire clearly existed. It also existed because it was clear even before they super-scaled that Amazon and Uber were doing something useful for which where existed a demand.

So far all we are seeing with the generative AI delusion is an exponentially exploding waste of resources in order to pollute my Youtube feed with slop. Every enterprise is trying "AI" and essentially all of them are finding it does not do what the people selling the tin claim it can.

There were no Amazon, or Uber or Internet evangelists trying to convince everyone that those things were useful or invent uses for them because there was no need: the value was obvious and real.

From your first linked article:

"There is no doubt we will see rapid advancements in some of the areas, but for others, like quantization, it is less clear. So while the cost of LLM inference will likely continue to decrease, its rate may slow down."

"Will LLM prices continue to decline at this rate? This is very hard to predict."

But go ahead and an assume massive declines in inferencing costs, we believe you. No doubt Altman and Musk will pass those savings onto the little man.

"or alternatively slows down a

What if we stopped having kids and concentrated on spiritual enlightenment, leaving the rich to deal with their own karma?

> It never ceases to amaze me how many people don't realize what's happening here. Even more so there are the people who realize it but just kind of put it out of their mind because the idea of the ultra wealthy dismantling capitalism is so far outside what people view as possible that they can't emotionally comprehend it even if they can understand it intellectually.

You act as if there's something we can do about it. The vast majority of the public *hate* all the AI garbage being packed into every piece of software and have been very vocal about it. The response of the ultra-wealthy is, "We know and we don't care because this is GOING to happen because we said so." Most people are more aware than you give them credit for, but we have a limited amount of energy we can put toward all the things happening in the world right now and we understand this is one of those thing

I think everyone realizes it that cares enough to consider it. Most don't bother thinking about it, at least beyond how they can make a buck. You are either a predator or a victim, the only solution is to not allow the game to be played. Sad thing is that most that imagine themselves predators don't realize they are the targets.

There is no post-AI economy, there is only homelessness, poverty and starvation. That's exactly what billionaires intend.

Then there will be an OpenAI coin and they will print their own net worth.

Current approaches to AI may need this, but what that says is they are doing it wrong. An "intelligence" is more than a neural network, no matter how big.

A human figures out what do to, an AI fakes it based on ample precedent it has been trained on. The two are not the same.