Silicon Valley has forgotten what normal people want

One of the most mortifying things about knowing a lot of techies is listening to them tell me excitedly about some very important discovery that they believe they have made. Recently, I ran into an acquaintance of mine, who began talking my ear off about an amazing discovery he’d made with LLMs. Knowledge, it turns out, is structured into language! You could put one word into ChatGPT and it might understand what you wanted, or make up a word and see if it understood what you meant! These amazing new tools have revealed that the English corpus contains so much about its speakers!

He concluded that LLMs are a discovery on par with writing.

Regular humans hit on this idea about a century ago; my most generous interpretation of what he was telling me was that he’d hit on a kind of naive, confused version of Structuralism; Saussure via a game of telephone. (There has been recent work on a similar point, which argues that one needs to understand LLMs via literary theory, but it starts with Saussure.) I tried to get out of the conversation as quickly as I could, not least because he seemed frustrated that I didn’t see things exactly as he did — a new behavior and likely a symptom of LLM overuse.

Not every discovery that’s new to you is actually new. For instance, there’s Elon Musk marvelling at the complexity of hands; I could point to a variety of disciplines for which this is 101-level stuff: artists, who have to figure out how to draw them; surgeons, who have to figure out how to operate on them; musicians and magicians, who rely on extremely fine motor skill to produce their work; neuroscientists and psychologists, who doubtless encountered the cortical homunculus early in their careers. Or Palmer Luckey claiming that “no one has done a postmortem” on the One Laptop Per Child computing project — because he didn’t know there’s a whole book about it called The Charisma Machine.

At its most absurd nadir, one is reminded of Juicero, a company that sold a $400 juicer that did the same work as squeezing its proprietary juice packs with one’s bare hands.

Look, discovering something that’s new to you is exciting — ask anyone who listened to me yell about the joys of European (higher-fat) butter — but you can’t take for granted that something that’s new to you is new to everyone. These things have in common a certain incuriosity that I have found endemic among a certain kind of tech enthusiast, particularly the ones who are most interested in startups and entrepreneurship. Perhaps they have been so siloed that they did not realize their “discovery” was well -known elsewhere, or perhaps their self-conception is that they are the smartest, and if they don’t know something, no one knows it.

There is a certain amount of hubris required to throw oneself at an unsolved problem — you have to believe you can solve it. But elsewhere, that hubris is a liability. It leads you to do weird things, like announce that Freud invented introspection and that it is a bonus that you simply do not engage in it.

When I think I have observed something important, my first impulse is to go to a library, or Wikipedia, or a person who I think may be knowledgeable, and see what else has been observed. For instance, when I had a concussion, I wanted to see if anyone else had written about what it was like to recover — the dry medical descriptions did very little for me. When I couldn’t easily find an account, I wrote my own. I still receive emails about it, years later, from people who are doing the same search I did, following their own concussions. But doing something like this requires you to take for granted that other people are smart, that smart people have always existed, and that very little in the human experience is new. That requires, you know, intellectual humility — and a willingness to think about other people’s experiences.

While this particular kind of hubris makes people crashing bores, it’s not just an annoying personal trait. It seems to have seeped into the professional side of Silicon Valley as well.

Within recent memory, people who made software and hardware understood their job was to serve their customer. It was to identify a need, and then fill it. But at some point following the financial crisis, would-be entrepreneurs got it into their heads that their job was to invent the future, and consumers’ job was to go along with that invented future. My guess is that they’re aping what they thought Steve Jobs was doing when he, for instance, got rid of the optical drives on the MacBook Air.

But Steve Jobs, famously, failed at inventing the future in the 1980s and got booted from Apple. We all know how things changed when he came back. But the iMac, the iPod, the iPhone were built with a need in mind. The iMac won because it was easy to use. The iPod was easier to take with you than a CD player and a stack of CDs. (It also was a way to play the MP3s you might have illegally downloaded.) The iPhone had the App Store, which expanded its utility well beyond any other mobile device.

Some of this was luck — introducing the right product at the right time. But each product offered consumers a distinct value proposition. Sure, early adopters jumped on each of these things because they were cool, but the uncool masses don’t care about that. They’ll buy something if it improves their life in a distinct way.

In the place of problem-solving technology, companies have jumped on successive bandwagons like NFTs, the metaverse, and large language models. What these all have in common is that they are not built to really solve a market problem. They are built to make VCs and companies rich. NFTs, like crypto, let VCs quickly unload investments with abbreviated lockup periods. The metaverse promised to enrich companies like Facebook by having people move all their socializing online, where it could be surveilled and monetized. In addition, Facebook’s metaverse required the purchase of hardware, which would then need regular upgrades.

At some point, our Silicon Valley overlords forgot that in order for their vision of the future to be adopted, people had to want it. That’s why NFTs, the metaverse, and the Oculus and Vision Pro never really found their customer base. AI is, admittedly, more useful — it’s good for organizing large swaths of data, for instance. LLMs have had widespread consumer adoption, at least as long as they remain free. But there is only really one customer for LLMs that can justify the massive cash incineration process that was required to build them: the US government.

There can only be a few winners on government contracts, though. So we are now treated to the spectacle of watching AI companies scramble. OpenAI is perhaps the funniest, because it is attempting to position itself as a consumer product.

Consider Sam Altman telling the world that he needed ChatGPT to tell him how to raise a baby. You exist. I exist. Our parents did not have LLMs, or even AI, and yet somehow we survived our childhoods, as did almost everyone else we knew growing up because childhood death rates in the US have been extraordinarily low — compared to most of the rest of human history — for decades. The technologies that allowed us all to survive our childhoods were sanitation, vaccines, and antibiotics. I would put money down that a mandatory measles vaccine will do more for the survival of American children than anything OpenAI has accomplished with all of its billions of dollars to date.

In any event, I presume what Altman actually did was hire a nanny.

Or consider Elon Musk telling us about our future humanoid robot servants. I have a robot servant. Several, actually: a dishwasher, a washer for my clothing, and a dryer. They aren’t very mobile, and yet they have saved me tremendous labor. My fridge is from the ’90s, and my microwave isn’t much younger, and both of those things have been remarkable in what they have done for me: made food storage and cooking easy, without AI involvement. It doesn’t seem like there’s much AI can do to improve things over the baseline that these machines have already established, especially since my “dumb” technology hasn’t required an update in more than 20 years. Saving money is valuable to me, too.

The people who tell us that AI will dominate our future and take our jobs are the people who are hoping that will be true. They may be hoping this because it makes them feel important, or because they want to be billionaires, or because they simply do not understand other people. I think that final point is underestimated. If you are going to provide me with a robot servant, I have a very clear bar: It’s gotta be at least as much bang for my buck as my dishwasher.

Normal people aren’t running around like chickens with their heads cut off, trying to automate every single part of their lives. Indeed, there are places in our lives where efficiency isn’t desirable. Vacation planning is sometimes suggested as a place AI can make our lives easier. For me, at least, planning the vacation is a pleasure in and of itself; it allows me to browse information about a place, consider what might be fun, and imagine myself doing it. If I have friends who have been to that place before, it gives me an excuse to talk to them, getting their recommendations. The entire process sharpens the anticipation I feel as the date for the vacation approaches. But if I wish to outsource that, I can do so already — that’s what cruise ships and theme parks are for.

LLMs are, at best, an enterprise technology that may make certain kinds of data organization easier, or coding faster. This has almost nothing to do with most people’s lives. Dinking around with code is a hobby many tech people enjoy and one the rest of us simply don’t care about. Making it easier to write code doesn’t change that I don’t want to write code. I have other hobbies!

The actual use for LLMs in most normal people’s lives is cheating on schoolwork. For adults, it’s looking up information — LLMs are in the process of supplanting Google Search. Google had been degrading its search project for some time, and the results just kept getting worse. This opened the door for an alternative, and the LLMs stepped through. How long that will last, I don’t know — the LLMs themselves will require money at some point and their frequently inaccurate (and sometimes plagiarized) results are killing the websites they rely on to generate information. Sure, it’s more inefficient to click through to a high-quality product, but how else do you plan to continue to have people generate high-quality information? No one has solved this problem.

Sometimes inefficiency is load-bearing. Take, for instance, the stock market. It is only open during certain hours, and only during certain days of the week. That means that during a panic, there is an artificial boundary that gives people time to calm down. This is effective; it’s one of the reasons that individual stocks sometimes undergo a trading halt during periods of hysteria. Now consider crypto, which is open for business 24/7/365: There is no way to pause a panic. One of the reasons the crashes in crypto are so huge and so fast is because there is no breaker to trip and no break in trading to allow traders to regroup. In fact, crypto panics are arguably exacerbated by the fact that many people literally cannot sleep because the market does not close.

There are other ways in which consumer AI is weird. Take the AI music apps, which are predicated on the idea that there are people in the world who want to make music but simply haven’t taken the time to learn how to play an instrument. There are likely very few of those people! Musicians aren’t bogarting creativity — they are people who enjoy making music. The rest of us just enjoy listening, which is an end in itself.

The place where AI music is most useful is for people who want to figure out how to get themselves onto Spotify playlists, accrue streams, and make money — that is to say, scammers. Similarly, the self-publishing market is rampant with AI slop, not because people are desperately trying to express themselves, but because it is easy to trick other people into buying slop on Amazon. And it’s not just the casual readers who get swindled, as demonstrated by the scandal around Shy Girl, the now-withdrawn novel that fooled Hachette. The end result for most people is that these AI tools make it harder for them to access art made by other people. And the end result for artists, of course, is that it’s harder to make a living.

How is it that all these wunderkinds trying to build the next product to take over the world haven’t thought about this? I think the answer is simple. They do not have much in common with normal people, and haven’t thought much about what normal people’s lives are like, or what normal people value. What they have been doing instead is getting high on their own supply — listening to VC podcasts, freaking themselves out about whether they’ll be able to keep up with AI agents, and otherwise getting increasingly more detached from reality.

I suspect this is how we wound up with NFTs, the metaverse, and the clunky VR/AR headsets. These are things that appeal to a very narrow set of people who are overrepresented in the VC and wannabe-tech-entrepreneur spaces. The Silicon Valley hype cycle worked overtime for those things, and I think we all know how this turned out. When was the last time you heard about a Bored Ape, or a Crypto Kitty, or any of the other novelties that briefly swept the nation? Did those novelties translate into a real, durable income stream for artists, musicians, and other creators, as we were promised? When was the last time you saw someone wearing Apple’s headset? Did Mark Zuckerberg’s Meta utopia ever develop legs?

Look, we all had a bit of fun at Marc Andreessen’s expense about his lack of introspection — but this is precisely the reason Silicon Valley keeps trying to force futures on consumers that they emphatically don’t want. A VC who is incapable of self-reflection will never notice that his bets on the future of consumerism are failing in exactly the same way every time. That VC hasn’t noticed, and indeed can’t notice, that his experience isn’t representative of what the ordinary person wants or needs.

Actually, while I’m picking on Andreessen, I want to point to a bit of that interview that didn’t go viral. It occurs right after the fatal introspection quote, but I think it gets to the real rot at the heart of Silicon Valley’s current culture. In it, Andreessen mentions psychedelics, saying he was discussing them with podcaster Andrew Huberman. “I was describing this phenomenon we see in Silicon Valley, where there are these guys who get under pressure, and they feel anxious or whatever, and someone tells them about psychedelics, and they try it,” Andreessen says. “And they kind of come out the other end as a changed person. They come out much more at peace, but then they tend to quit their companies.”

In Andreessen’s telling, Huberman suggests that these people may be happier, and better off. And Andreessen says, “Yeah, but their company is failing.”

The hubristic entrepreneurs (and the VCs who need them) are a relatively small slice of the population. The majority of us would much rather be happy than try to found a company that takes over the world — sacrificing the majority of our waking hours, our hobbies, and likely many of our relationships in the process. It may be the case that the real way to shape the future isn’t to dictate it to consumers. It is simpler just to give people things they actually want.