The AI dooooooom thread

[Sheilbh]

Saw this post on reddit, summarizing the drivers of the upcoming AI bubble bursting. It references a podcast (which I didn't listen to at all). It seemed reasonable enough to me, so I'm posting it here for people to tell me how it's not 

I don't know if I can.

Having looked at lots of AI products and worked on them being used in businesses I do not think there is a business model (yet) for companies like OpenAI. I can't work out what it's for.

The bit of AI that I think is going to take off and grow is basically the SaaS piece that is an additional enterprise license/subscription. So, for example, Salesforce, Oracle, your code repository. Those sort of environments where there is a lot of data necessary for your business and the AI improves or transforms functionalities I can absolutely see the business case for.

I'm not 100% sure where Gemini sits within this frame yet (or Microsoft CoPilot) but my instinct on Google is they are very much in a throw it all around and see what stick mode.

However in terms of what happens I don't want to be too alarmist (and there may well be positives)...but, there has been massive investment in this (both the phsysical piece and the models). The goal being that it is transformative from a productivity perspective, allowing companies to earn more with less labour largely by being able to automate (largely) white collar work. If that is true it will have enormous economic and social implications that will need to be worked through.

If it isn't true and doesn't work out, then I think in terms of scale it will be one of the biggest misallocations of capital ever. The fallout from that will have enormous economic and social knock on effects and implications too.

[Josquius]

I am seeing various reports around about corporate attempts to switch to enterprise AI which are 95% failures.
Not sure on how reliable this is and how they're defining failure.
But certainly there are signs things are imploding.... And much earlier than anticipated.
One explanation I heard was the AI is proving incapable of "learning" specific business processes.

I can definitely see some business uses in helping individuals with writing, sorting/summarising big data sets (not to be relied on without doing it again yourself but... It's a good first run) , and most importantly transcribing and summarising meetings (with maybe 80% accuracy? There's some fun developments at work on this I can't really talk about).
But this sounds like it's replacing personal assistants and secretaries....which....already normal workers don't have those and the people that do are generally powerful enough they can insist on a human.

[crazy canuck]

Saw this post on reddit, summarizing the drivers of the upcoming AI bubble bursting. It references a podcast (which I didn't listen to at all). It seemed reasonable enough to me, so I'm posting it here for people to tell me how it's not 

Better Offline covers this most thoroughly. The gist is this:

AI companies are spending tens and hundreds of billions of dollars

AI companies are earning (single digit) billions of dollars

AI companies are profiting no dollars, or negative billions depending on your position on "investment in infrastructure")

AI companies almost all rely on other AI companies to use their product, generating the revenue they do make from their hardware off contracts from the companies that write chatbots and other programs that have guaranteed token use

NVidia, valued at over $4 Trillion, relies on continued sales of their chips to the hardware folks

Nearly all the AI companies are not traded publically, yet are valued in the tens-to-hundreds of billions, making them nearly impossible for someone to buy

When the non-profitable software guys stop paying the hardware guys, NVDA has a dip in stock, a "correction," and it'll scare the shit out of the markets

Those startup companies that were too expensive to buy will now be left with no more venture capital and huge negative run rates, meaning less NVDA buys, meaning more dip, more VC pull-back, fewer and less buys, etc, etc, etc

Because they are all incestuous with each others equipment and software, when one falls, it'll rip the others like crabs in a pot.

https://www.reddit.com/r/technology/comments/1mx1p2c/the_warning_signs_the_ai_bubble_is_about_to_burst/

Yeah, this reminds me a lot of the late 1990s.  Lots of money going in to try to leverage a new technology in ways that made no economic sense.

[Jacob]

It Took Many Years And Billions Of Dollars, But Microsoft Finally Invented A Calculator That Is Wrong Sometimes

It's not AI winter just yet, though there is a distinct chill in the air. Meta is shaking up and downsizing its artificial intelligence division. A new report out of MIT finds that 95 percent of companies' generative AI programs have failed to earn any profit whatsoever. Tech stocks tanked Tuesday, regarding broader fears that this bubble may have swelled about as large as it can go. Surely, there will be no wider repercussions for normal people if and when Nvidia, currently propping up the market like a load-bearing matchstick, finally runs out of fake companies to sell chips to. But getting in under the wire, before we're all bartering gas in the desert and people who can read become the priestly caste, is Microsoft, with the single most "Who asked for this?" application of AI I've seen yet: They're jamming it into Excel.

Excel! The spreadsheet program! The one that is already very good at what it does, which is calculation and data analysis. You put some numbers in and it spits some numbers out. According to The Verge, "Microsoft Excel is testing a new AI-powered function that can automatically fill cells in your spreadsheets." Using natural language, the idea goes, you tell it what you want and then the AI will "classify information, generate summaries, create tables, and more."

...

https://defector.com/it-took-many-years-and-billions-of-dollars-but-microsoft-finally-invented-a-calculator-that-is-wrong-sometimes

[crazy canuck]

At this point in the Trump presidency making numbers up is a virtue

[Valmy]

Saw this post on reddit, summarizing the drivers of the upcoming AI bubble bursting. It references a podcast (which I didn't listen to at all). It seemed reasonable enough to me, so I'm posting it here for people to tell me how it's not 

Better Offline covers this most thoroughly. The gist is this:

AI companies are spending tens and hundreds of billions of dollars

AI companies are earning (single digit) billions of dollars

AI companies are profiting no dollars, or negative billions depending on your position on "investment in infrastructure")

AI companies almost all rely on other AI companies to use their product, generating the revenue they do make from their hardware off contracts from the companies that write chatbots and other programs that have guaranteed token use

NVidia, valued at over $4 Trillion, relies on continued sales of their chips to the hardware folks

Nearly all the AI companies are not traded publically, yet are valued in the tens-to-hundreds of billions, making them nearly impossible for someone to buy

When the non-profitable software guys stop paying the hardware guys, NVDA has a dip in stock, a "correction," and it'll scare the shit out of the markets

Those startup companies that were too expensive to buy will now be left with no more venture capital and huge negative run rates, meaning less NVDA buys, meaning more dip, more VC pull-back, fewer and less buys, etc, etc, etc

Because they are all incestuous with each others equipment and software, when one falls, it'll rip the others like crabs in a pot.

https://www.reddit.com/r/technology/comments/1mx1p2c/the_warning_signs_the_ai_bubble_is_about_to_burst/

This is the thing that really baffles me about 21st century economics. There seems to be an infinite amount of money to throw at things that have zero obvious potential to turn a profit. While vital services and goods people desperately need and want to buy are allowed to just wither and die, looted by private equity firms, and nothing seems to be coming up to replace them. At least not quickly. It is crazy town out here.

[crazy canuck]

Welcome to late stage capitalism Valmy

[Sheilbh]

This is the thing that really baffles me about 21st century economics. There seems to be an infinite amount of money to throw at things that have zero obvious potential to turn a profit. While vital services and goods people desperately need and want to buy are allowed to just wither and die, looted by private equity firms, and nothing seems to be coming up to replace them. At least not quickly. It is crazy town out here.

I'm not sure on that with AI.

This requires physical infrastructure. There are huge data centres being built and significant investments in power generation for them, not to mention the chips industry which is real but also significant from a national security perspective. But I think it is real capital expenditure and the purpose is capital deepening. The big bet on this is that it will have significant increases in productivity. I agree I can't see a route to profit for any of the standalone products yet, but I can in other areas. If that bet pays off (obviously, as I say, transformative social and economic effects too).

But I think (and this could be bullshit as I'm just thinking this through) part of the dynamic and infinite cash is that the assumption is it'll go like the rest of big tech and consolidate into close to a monopoly. So if this works and you're the one with the model that "wins", there'll be huge market power (and profit). What's interesting is that every big tech company feels the need to throw everything at it because the potential prize is so big. And I don't think that's really happened before - I don't think everyone felt the need to build their own browser, or search engine etc.

And I mentioned it with the chips and the geopolitical angle - but I think that is also important. From what I understand it looks like China has been less impeded by US export restrictions than hoped. It is one of the strategic sectors China's leaders have identified for the next stage of their development and they are aspiring to self-sufficiency and then international competitiveness. As a general tool Deepseek is very impressive. There is opacity on how and how deeply China is embedding AI but from what I've read it is booming on the consumer side and absolutely getting embedded in day-to-day for business. I can't help but think of Jake's comment about the level of security theatre in China recently as I'm sure a lot of that is security paranoia - but also as China's economy, how much of it is just a make-work scheme for people who are not going to be getting the lower cost manufacturing jobs China's moving out of? And could we something similar in other sectors?

[Sheilbh]

I'd also add the bits of the MIT report are interesting but I think some of the way it's been discussed are a little misleading.

It was based on interviews and surveys around this and found that about 5% achieved quick revenue growth. They said some big companies and some startups are getting really positive results "because they pick one pain point, execute well, and partner smartly with companies who use their tools".

They say the issue is a "learning gap" for both the actual AIs and also organisations using them. So execs blame regulation or the model's performance, but the report suggests it's actual implementation that's the issue. So ChatGPT works for individuals but doesn't work at a business level because it's not learning from workflows (can confirm based on experiments I've seen - ChatGPT is the least useful and in almost every area more specific tools are better).

Similarly it finds that about half the AI budgets in companies are going to sales and marketing tools - which I think would have a really bad ROI on so many levels ("please genericise my copy"). The best return is back-office automation.

I think a lot of this is probably true of basically every piece of new tech hitting companies. It totally lines up with my experience and is why I lean towards the view that the "winners" will be the Oracles and the Salesforces etc. I also think it's why Sam Altman is out there posting things about the Manhattan Project and the Death Star because that sort of existential threat angle is one hell of a marketing tool when, at this stage, you're quite possibly just building a plug-in enterpise SaaS functionality. "Look on my works ye mighty and despair", said the creator of Trello.