i feel like this is going to play out very similarly to voice assistants. there was a huge boom in ASR research, the products got a lot of hype, and they actually sold decently (at least alexa did). but 10 years on, they’ve been a massive failure, costing way more than they ever made back. even if ppl do think chatbot search engines are exciting and cool, it’s not going to bring in more users or sell more products, and in the end it will just be a financial loss
(Responding to this a week late)
I don’t know much about the history of voice assistants. Are there any articles you recommend on the topic? Sounds interesting.
But anyway, yeah… this is why I don’t have a strong sense of how widespread/popular these “generative AI” products will be a year or two from now. Or even five years from now.
(Ten years from now? Maybe we can trust the verdict will be in at that point… but the tech landscape of 2033 is going to be so different from ours that the question “did ‘generative AI’ take off or not?” will no doubt sound quaint and irrelevant.)
Remember when self-driving cars were supposed to be right around the corner? Lots of people took this imminent self-driving future seriously.
And I looked at it, and thought “I don’t get it, this problem seems way harder than people are giving it credit for. And these companies show no signs of having discovered some clever proprietary way forward.” If people asked me about it, that’s what I would say.
But even if I was sure that self-driving cars wouldn’t arrive on schedule, that didn’t give me much insight into the fate of “self-driving cars,” the tech sector meme. It wasn’t like there was some specific deadline, and when we crossed it everyone was going to look up and say “oh, I guess that didn’t work, time to stop investing.”
The influx of capital – and everything downstream from it, the trusting news stories, the prominence of the “self-driving car future” in the public mind, the seriousness which it was talked about – these things went on, heedless of anything except their own mysterious internal logic.
They went on until … what? The pandemic, probably? I actually still don’t know.
In 2018 analysts put the market value of Waymo LLC, then a subsidiary of Alphabet Inc., at $175 billion. Its most recent funding round gave the company an estimated valuation of $30 billion, roughly the same as Cruise. Aurora Innovation Inc., a startup co-founded by Chris Urmson, Google’s former autonomous-vehicle chief, has lost more than 85% since last year [i.e. 2021] and is now worth less than $3 billion. This September a leaked memo from Urmson summed up Aurora’s cash-flow struggles and suggested it might have to sell out to a larger company. Many of the industry’s most promising efforts have met the same fate in recent years, including Drive.ai, Voyage, Zoox, and Uber’s self-driving division. “Long term, I think we will have autonomous vehicles that you and I can buy,” says Mike Ramsey, an analyst at market researcher Gartner Inc. “But we’re going to be old.”
Whatever killed the “self-driving car” meme, though, it wasn’t some newly definitive article of proof that the underlying ideas were flawed. The ideas never made sense in the first place. The phenomenon was not really about the ideas making sense.
Some investors – with enough capital, between them, to exert noticable distortionary effects on entire business sectors – decided that “self-driving cars” were, like, A Thing now. And so they were, for a number of years. Huge numbers of people worked very hard trying to make “self-driving cars” into a viable product. They were paid very well to do. Talent was diverted away from other projects, en masse, into this effort. This went on as long as the investors felt like sustaining it, and they were in no danger of running out of money.
Often the “tech sector” feels less like a product of free-market incentives than it does like a massive, weird, and opaque public works product, orchestrated by eccentrics like Masayoshi Son, and ultimately organized according to the aesthetic proclivities and changing moods of its architects, not for the purpose of “doing business” in the conventional sense.
Gig economy delivery apps (Uber Eats, Doordash, etc.) have been ubiquitous for years, and have reported huge losses in every one of those years.
Which brings us to the question - what is the point of all this? These platforms are all losing money. Just think of all the meetings and lines of code and phone calls to make all of these nefarious things happen which just continue to bleed money. Why go through all this trouble?
Grubhub just lost $33 million on $360 million of revenue in Q1.
Doordash reportedly lost an insane $450 million off $900 million in revenue in 2019 (which does make me wonder if my dream of a decentralized network of pizza arbitrageurs does exist).
Uber Eats is Uber’s “most profitable division” 😂😂. Uber Eats lost $461 million in Q4 2019 off of revenue of $734 million. Sometimes I need to write this out to remind myself. Uber Eats spent $1.2 billion to make $734 million. In one quarter.
DoorDash’s total orders grew 27% to 467 million in the fourth quarter. That beat Wall Street’s forecast of 459 million, according to analysts polled by FactSet. Fourth quarter revenue jumped 40% to $1.82 billion, also ahead of analysts’ forecast of $1.77 billion.
But profits remain elusive for the 10-year-old company. DoorDash said its net loss widened to $640 million, or $1.65 per share, in the fourth quarter as it expanded into new categories and integrated Wolt into its operations.
Do their investors really believe these companies are going somewhere, and just taking their time to get there? Or is this more like a subsidy? The lost money (a predictable loss in the long term) merely the price paid for a desired good – for an intoxicating exercise of godlike power, for the chance to reshape reality to one’s whims on a large scale – collapsing the usual boundary between self and outside, dream and reality? "The gig economy is A Thing, now,” you say, and wave your hand – and so it is.
Some people would pay a lot of money to be a god, I would think.
Anyway, “generative AI” is A Thing now. It wasn’t A Thing a year ago, but now it is. How long will it remain one? The best I can say is: as long as the gods are feeling it.
I think the key behind a lot of these “AI” driven technologies, including self-driving cars and automated chatbots is that they’re solving problems that are solvable maybe 95-98% of the way, but the final 2% is much, much harder than you might expect based on the first 98%.
Like, it’s been years since the first “self-driving” car that could drive through a test course without human intervention, but we’re still far from “and a thousand of them could do this on the highway every day and not one would randomly slam on the brakes and cause a pileup”.
But much of the tech industry isn’t actually driven by people trying to change the world, it’s driven by people trying to make lots of money, and solving 95% of a problem is often enough to convince investors to give you money. And then your company becomes a sort of long, slow pyramid-scheme - everyone who invests in it waits for people to get even more psyched, and then sells their investment at a profit.
In this way, DoorDash doesn’t *need* to make a profit for its investors to make a profit - it just needs to beat expectations on enough metrics that the current investors can convince other investors to buy their shares for more than they originally cost.
Eventually the bubble has to burst, because that missing 2% just isn’t there - DoorDash will probably never be profitable, Tesla won’t have truly reliable self-driving cars for decades, after enough publicly visible disasters caused by Bing giving batshit advice people will stop trusting AI “search engines”, etc. And it’ll suck for whoever’s invested in them at the very end, although they’ll probably be making enough money from doing the same thing with the next big fad that they won’t care that much.
It’s basically the Gell-Mann amnesia effect that’s driving the hype. When you see an AI tasked with something in your expertise, it’s easy to spot the 2% that’s missing. But for any other task — even ones you have “95% familiarity” with — the issues aren’t apparent. It can even be the exact same program generating both responses, and we can still forget to carry over our skepticism from the first response to the second.
For a dose of cold water, every investor involved in any AI-related project needs to try using ChatGPT to generate investment advice.
