The emperor's G-string

The future is going to be messy.

And I'm being uncharacteristically optimistic in saying that.

It might not be what you think it is. The biggest danger to the continued survival of humans on the planet will continue to be humans. Some of the people hyping the near inevitability of an AI apocalypse are selling you solutions and systems associated with that product. They want to impress you with the product's efficacy by framing it in the classic problem/solution marketing argument.

Some of the other people hyping the AI overlord hypothesis might be academics or industry nerds who are currently basking in the limelight for the first time. This may have had a less than salubrious impact on their truthiness. Imagine being a respected academic and/or data scientist, toiling away for decades in relative obscurity. Suddenly you find yourself a household name. You're being interviewed, your face is plastered on every screen in the mediasphere. Animals are biologically programmed to recognize and work within status systems, and the smartest people on the planet also happen to be . . . you guessed it, animals.

One salient feature of the first group of people is that they have convinced venture capitalists, governments, and anybody with the ability to borrow money to invest in infrastructure. To the tune of $1.7 – 7 trillion.

Now I know you're thinking, "that's a super wide range, what are your error bands?" And rightly so, but which report would you like to believe, OECD? McKinsey? Bloomberg? Any government on earth?

Let's just say that the top of that range exceeds the annual GDP of every country on earth except the US and China (the bottom of it sits around the 12th–15th largest — roughly Spain or Australia), and all of it has been or will be spent on compute infrastructure. Since you're here, I'll assume you can factcheck that statement to your heart's content.

That's trillions of dollars that is not being spent on other things. Like food for poor people, bullets, and other directly usable tools for killing other humans, or people with bullshit jobs¹.

The last one is particularly tricky, to put it mildly. There are millions of people working jobs to support bureaucracies that have bloated out of control. Many of these jobs are mind-numbingly painful, but they pay for braces on kids teeth, they pay for Netflix and mortgages on suburban houses. And they represent a non-trivial part of the tax base as a significant portion of the aforementioned GDP. In short, you could argue that they aren't necessary, but there are real human and administrative challenges to replacing those jobs with systems.

Now, many of those jobs could have been automated a long time ago with better workflow design and a smidgen of good old machine learning. However, if you are the boss of 45 people who have those jobs, if you are savvy enough to recognize the potential for improvement, you are also sufficiently cautious to steer your organization well away from that course of action. Talking yourself out of a job has traditionally been seen as a lousy career move. No one is more aware of that, at a cellular level of self-preservation, than a person who has a management career in an organization large enough to warrant a critical mass of bullshit jobs.

OK, this is the part of the post where I will try and get helpful. What do I think will help?

First let's set some known conditions, our priors.

The money is spent. It's not coming back. No putting the genie back into the bottle. The organizations that are doing this have government-equivalent power to borrow, allocate, exchange, and oxidize money. So, they did. And they will continue to do so.

There's a likelihood of an overall net benefit from the buildout itself, but not before a degree of discomfort that is not calculatable at the moment, at least not by me. In the not too distant future industrial failure aficionados will be spelunking in abandoned data centers, wondering how the people who built them could be so stupid. That being said, I think there's a good chance that the buildup of all that compute, in and of itself, will have a positive effect in our data and business world. When the pain subsides, we could be left with cheaper, more readily available, and more regional compute, especially if it is built to be upgradeable.

Everyone who is not currently using LLMs will be using them within five years at the most. They will be built into everything, even (maybe especially) if they are not fit to purpose.

The number of people who actually know how to use those systems to produce consistently useful results will be slightly greater or equal to the number of people who knew how to use Boolean operators with Google twenty years ago. That is to say that the number will increase, but we will never reach a point where the expert use of these systems is generally, or genuinely, ubiquitous. Let's call these folks power users.

Which leads me to the next point: the systems are currently being sold as something you don't need any training to use. Not if you are just a user. You don't need any training if the goal of said use is to burn tokens, generating profits for the companies that provide the systems to you. The control surfaces and interfaces are cleverly designed to do just that. After all, just servicing the debt on data centers is astronomically expensive.

The ratio between these two groups, power users and users, will represent the likelihood that an organization will successfully navigate the next few years of upheaval: I'm not trying to be overdramatic, but we're all in for a rough ride. Some people have saddles.

Neither of these cohorts is homogeneous. The power users are people who are interested enough to educate themselves and stay informed in order to make the best use out of the tools they use every day. They have become power users, distinct from experts by virtue of their relative inability to create or work on a project to create their own LLM and characterized by their skill at setting them up and fine tuning them to get the best out of the system. They are system thinkers, they are comfortable with the tech stack, and they know how to navigate and make use of documentation. They come in all shapes and sizes, but these are the behavioral traits they share. They are a little bit obsessed with figuring things out, taking things apart, putting them back together. They are tinkerers, do-it-yourselfers, and makers.

They are the emperor's G-string. Along with the real experts, they will cover the embarrassingly naughty bits of the fails to come.

Users are, well, everybody else. People who have the patience to wait for shared service centers to answer their questions. Employees with advanced degrees in ticket opening, social engineering, and when all else fails, flat-out whining. Many of them view the systems they work with every day as a black box. They live quite comfortably in that space, where other nice people are available to fix things. They go through the drive through at Starbucks instead of buying their own coffee beans and espresso machine. They have an air fryer instead of a Japanese chef's knife. Or they just order takeout. They operate in a world where stuff just works, and if it doesn't you replace it. They are obsessed by other things. They shop for fun. They like reality TV.

Please don't misunderstand me, I belong to (a very small non-expert subset of) the first group, but I don't mean to be pejorative about the second. They are the overwhelming majority and taken as individuals they are much better at certain types of human behavior than the power users. And they are as heterogeneous as the power users, maybe even more so. They might be so into gardening that they are collecting legacy seeds and testing soil Ph. They might have an encyclopedic knowledge of competitive team sports or Victorian novels.

If you've read this far, you either know me, or perhaps you're asking yourself, "who is this guy"? I'm a full-blown reading addict. I'm late middle-aged. I'm an expatriate. I'm technical. I'm an overconfident writer, making grandiose, even sweeping statements. I have a propensity to use words I feel are very specific and accurate that some people may see as ostentatious displays of poorly applied erudition, which is fun for me. I've had several careers, most of which in the past two decades revolved around practical linguistics: language training, translation, academic and commercial copyediting, and post-editing.

I have a polymorphously perverse knowledge base, I like to wave it incessantly in the wind.

¹ In the sense of the book with the same title by the late anthropologist David Graeber, ISBN 978-1-5011-4331-1