GenAI FOMO has spurred businesses to light nearly $40B on fire

Agreed. Smug dismissal of new ideas is such a lazy shortcut to trying to look smart. I'd much rather talk to someone enthusiastic about something than someone who does nothing but sit there and say "this thing sucks" every time something happens even if person #2 is incidentally right a lot of the time.

Similarly, VC sets barrels of money on fire.

And depending on how you look at it, science itself is experimentation, but at least it mostly results in publications in the end, that may or may not be read, but at least serve as records of areas explored.

You put my thoughts into words better than I could.

This reminds me of the exploration-exploration trade-off in reinforcement learning: you want to maximise your long term profits but, since your knowledge is incomplete, you must acquire new knowledge, which companies do by trying stuff. Prematurely dismissing GenAI could mean missing out on new efficiencies, which take time to be identified.

Or (4), because a sales team convinced them there was some magic wand they could buy that would triple their productivity.

You're giving company executives way too much credit in general. I'm sure there are unicorns out there where conscientious stewards of the company's long-term health are making measured, rational choices that may pay off in a decade, but it's a tiny minority of companies. Most are run by narcissistic short-term corporate raiders whose first priority is looting the company for their own profit and second priority is cosplaying as a once-in-a-generation genius "thought leader" in a never-ending battle to beat away the nagging (and correct) thought that they're nepo-babies who have no clue what they're doing. These morons are burning money because they are stupid and/or because it benefits them and their buddies in the short-term. They burned money on blockchain bullshit, they burned money on Web 2.0 bullshit, they are burning money on AI, and they will be burning money on the next fad too. The fact that AI might actually turn out to be something real is complete serendipity; it has nothing to do with their insight or foresight. The only reason they ever look smart is because they're great at taking credit for every win, everyone else immediately forgets all their losses, and op-ed writers and internet simps all compete to write the most sycophantic adulations of their brilliance. They could start finger-painting their office windows with their own feces and the Wall Street Journal would pump out op-eds saying "Here's why smearing poop on your windows is actually a Really Brilliant Business Move made by Really Good-Looking Business Winners!" Just go back and re-read your comment but think "blockchain" instead of "AI" and you'll see clearly how silly and sycophantic it really is.

> Well run large companies

I've never heard of such a thing

My monkey jpegs are surely going back up! Just have to keep hodling.

> has exactly one purpose: prevent any single entity from controlling the contents.

I'd like to propose a different characterization: "Blockchain" is when you want unrestricted membership and participation.

Allowing anybody to spin up any number of new nodes they desire us the fundamental requirement which causes a cascade of other design decisions and feedback systems. (Mining, proof-of-X, etc.)

In contrast, deterring one entity from taking over can also be done with a regular distributed database, where the nodes--and which entities operate them--are determined in advance.

let's use the term xype for anything being hyped without a good reason .. :-)

> No one knew what “blockchain” was, how it worked, or what could be used for.

Not the blockchain itself, but the concept of an immutable, append only, tamper-proof ledger underpinning it is a very useful one in many contexts where the question of authenticity of datasets arises – the blockchain has given us a ledger database.

The ledger database is more than just a hash as it also provides a cryptographic proof that the data is authentic via hash chaining, no ability to delete or modify a record, and the entire change history of any record. All these properties make the ledger databases very useful in many contexts, especially the ones where official documents are involved.

I suppose my question that I’m too tired to napkin math on is if those benefits justify the sometimes astronomical cost of the tooling or integration

But that's only the company side and not the customer side.

Klarna also cut costs replacing support with AI. Didn't work well so ha to rehire.

What is missing even in this article is the install and expected failure rate of the dominant GB300 servers. Numbers I heard were, "~15% annual failure and it's not worth trying to swap/repair". That means in 5 years these entire installs are down more than half. Of course they can install the NEW GX500turbo servers which are 4x the compute, but 2x more power hungry. How much will that cost? What is the hyperscaler write down ~$200B/yr? Better have some income to make that up. They've got only 3 years to get there.

That still means All New data centers. They aren't being built for for this now, and so the old ones'll have to get ripped out and rebuilt (in place?) before they get the new servers. I do think they've planned the external power delivery, but not cooling or IP infra. It's a CF.

But business process outsourcing companies actually perform the services they get paid for, so it doesn't make sense to compare them to GenAI.

BPO is a growing industry that isn't shrinking call centers have expanded as well.

If it was true these were replacing anything it would be very clear in those sectors and it isn't. The real effect of end to end automation from LLMs is small to negligible. The entire "boring" industry is still chunging along and growing as it did before.

> Article doesn't understand that cutting "between five and 20 percent of support and admin processing" is really valuable, instead it seems to want to dismiss that as a dull failure.

To whom? From the customer perspective, it sounds like a shittier level of service is coming, which is a kind of failure.

I can just speak for myself, but my F500 has been setting money on fire chasing AI. Truly terrible ideas are being pursued just for the sake of movement. Were “AI” not part of the pitch, it would have been immediately tossed in the bin as a waste of time.

Ideas which are not terrible, instead have awful ROIs. Nobody has a use case beyond generating text, so lots of ideas about automating some text generation in certain niches. Not appreciating that those bits represent 0.1% of the business ventures. Yet they are technically feasible, so full steam ahead.

Do you have reason to believe that the MIT researchers didn't interview who they say they did? Or that those people don't have the credentials the researchers claim they do?

It's possible the study is flawed, or is more limited than the claims being made. But some evidence is necessary to get there.

I imagine it is the same, but 2000 was hardly a dead end. So there will be lots burned but keep on at it as the tech will revolutionize the world.

I would say a bit more like 1997. I was young but kept thinking, "That which is good is not novel, and that which is novel isn't good." That said, we had the giant crash, but based on the dow jones the low in 2002 was the same as August 1998 but you can really look at the lost decade as to the true impact of the bubble.

The question isn't if there will be a crash - there will - but there are always crashes. And there are always recoveries. It's all about "how long." And what happens to the companies that blow the money and then find out they can't fire all their white collar workers?

(Or, what happens if they find out they can?)

Truly, except now everyone's online and spending time online and spending cash online.

So you're saying they're basically right, but 5 or 10 years too optimistic on the timeline?

It'll be interesting to see what sort of Chewy-equivalents emerge after the pets.com-analogues collapse.

1999 was triggered by fraud, there was far more demand on paper for high bandwidth applications than was justified for all the fiber layout and over-investment in Internet technologies.

The difference today is that every piece of capitalism immediately 100% utilized once it is plugged in.

It's a bubble if the tech or audience wasn't there and capabilities weren't improving weekly.

There's definitely people who don't understand the tech talking about applying it increasing the failure rate of software projects.

> Right now VC FOMO is the only thing stopping all that from blowing up

That was probably 2-3 years ago.

I'd be surprised if VCs hadn't already figured out they're in a bubble. They're probably playing a game of chicken to see who will remain to capture the actually profitable use cases.

There's also a number of lawsuits going on against AI companies regarding piracy and copyright. It's already an established fact in the courts that these companies have downloaded ebooks, music, videos, and images illegally to train their models.

It'll certainly be lonely on HN without pages of AI articles to read daily..

/s

CEOs are just salivating at the prospect of firing most of their staff and replacing them with AIs. And then you have hype men at these AI companies blowing a burning coal mine of smoke up their asses. And every month the products become even more expensive for even more incremental gains. The hype is unsustainable.

How many "we will have AGI by X/X/201X" predictions have we blown past already?

The VCs are the small players in this bubble now. Big traditional finance is helping Microsoft, Google, Meta, and Amazon build out their datacenters, and the infrastructure to power them. VCs have a lot of money compared to your startup, but they can’t finance a trillion dollars in construction projects.

They’re all so highly levered up that they can’t afford for the bubble to pop. If this goes on for another couple of years before the pop, we may see “too big to fail” wheeled out to justify a bailout of Google or Microsoft.

I like your phrases "a smoking crater in the ground" and "a dumpster fire burning next to a gasoline truck" enough to steal them for use in my own critiques.

Would love to hear more (private is OK)

Moreover, money is not destroyed by spending, but rather circulates. Right now it's flowing around faster -- a manic phase for a portion of the economy.

In other eras, everyone got excited and went to tent revivals.

Granted there could be an opportunity cost -- the real effort and electricity could be used elsewhere -- but only if it were possible to create a similar amount of excitement about something useful, like putting solar panels everywhere. But that takes different people and different skills, so maybe this costs nothing?

(Money can be created and destroyed -- it doesn't just circulate -- but that destruction happens when loans are repaid within a fractional reserve system. Which is kind of a scam, but money itself is, so, whatever.)

Once you internalize that this is all sort of a scam, does that change your behavior? Maybe you start making NFTs or minting shitcoins.

> I also think things like "is this chest Xray cancer?" are going to be hugely impactful.

Yes, but https://radiologybusiness.com/topics/artificial-intelligence...

Nine years ago, scientist Geoffrey Hinton famously said, “People should stop training radiologists now,” believing it was “completely obvious” AI would outperform human rads within five years.

I've been quite happy with thinking of agentic IDE operation as being akin to a highly energetic intern. It's prone to spiraling off into the weeds, makes silly mistakes, occasionally mangles whole repos (commit early, and often), and needs very crisp instruction and guidance. That said, I get my answers back in minutes/hours rather than days/weeks. For the cost, for things that would otherwise be delivered by an intern or college-hire SDE, it's a pretty solid value vs. paying a salary and keeping a desk available.

What it isn't, at present, is an investment in the future. I'm not making these virtual interns better coders, more thoughtful about architecture, or more autonomous in the future. Those aspects of development of new hires are vastly more valuable than the code output I'm getting in my IDE. So I'm hoping that we land in a place where we're fostering both rather than hoping that someone else is going to do the hard work of closing the agentic coding gap and growing maturity. Pulling an Indiana Jones style swap could be a really destructive move if we try to pull the human pipeline out of the system too early.

Just paying attention to near term savings runs a real risk falling into that trap.

Agree with this, the "find this thing in my spaghetti codebase" is far and away the best use of LLMs I've seen. Fill in the rest of this switch statement, populate this struct from this database call, etc. also work pretty well. I would love if I could get a small model that ran locally that was able to pull off those 2 tricks. Explaining code works sometimes, but even the biggest models are still prone to getting confused and/or making stuff up that isn't there. I don't like the agentic features at all and expect these to mostly die because they're expensive and, IMO, only provide the illusion of productivity.

The market can stay irrational longer than the investor can stay solvent.

"The market can remain irrational longer than you can remain insolvent." --John Maynard Keynes

Lots of people lost their shirts shorting the housing market prior to the 2008 crash. (_The Big Short_ highlights those who were successful, but plenty of people weren't.) But it was undoubtedly a bubble and there was a world-wide recession when it popped.