I note that Musk is not particularly good at speaking in public, but that he is fully himself, which makes him free in a way Zuckerberg is not. All of Musk’s bottomless flaws are right there on the surface for everyone to see, which is why millions of people hate his guts, and also why millions of people get to feel like personal stakeholders in his dreams. Whereas Zuckerberg is an easier mark, Musk’s lack of concealment paradoxically makes him harder to hit. It somehow makes sense that Musk is the man who builds rocket ships to escape Earth, whereas there has always been something dissonant in Zuckerberg as the man who connected billions of individuals in a web of their own self-expression.

Surprisingly insightful, oddly sad.

Our technology is increasingly treating the territory between the query and the answer as dead space to be eliminated, when that territory is where most of the learning actually happens. The danger is not that people will stop asking questions. It is that questions will become endpoints. The loss is not serious in any single case. But fewer detours and fewer unexpected discoveries will have a cumulative effect. Over time, people trained this way become better at extracting ready-made conclusions than building connections of their own.

Another example of the cost of optimisation.

With all that has happened in the United States in recent years, it’s out of touch. The industry as a whole doesn’t even pretend anymore to be making the world a better place, it’s just high on technical prowess. For a company that purportedly positions itself at the intersection of humanity and technology, it’s a polished, hollow, repetitive spectacle.

Suppose you could travel to Egypt in 3000 BC and copy, in flawless hieroglyphics, the contents of every temple library, every architectural plan, every priestly manual, every commercial ledger. Then suppose you travelled to Mesopotamia and did the same in cuneiform. Consolidate everything you could find in the languages of that era, and then proceed to train a large language model on it. Full transformer architecture, self-attention, the whole enchilada.

The result would be a system capable of a certain kind of intelligence. It could predict floods from astronomical cycles. It could draft administrative correspondence. It could generate plausible religious commentary. But it would have no capacity for what the Greeks would later call the syllogism. It would carry no trace of Roman legal abstraction, and have no conception of the empirical method that wouldn’t emerge for another four millennia.

The claim is that the intelligence of a large language model isn’t in its architecture — it’s in the pre-existing “social complexity of the civilisation whose language it digested.” An LLM can record and approximate the complexity, but it is stuck there.

Jaron Lanier argued similarly a couple of years ago:

The new programs mash up work done by human minds. What’s innovative is that the mashup process has become guided and constrained, so that the results are usable and often striking. This is a significant achievement and worth celebrating—but it can be thought of as illuminating previously hidden concordances between human creations, rather than as the invention of a new mind.

What this study found is that while at the individual level it might seem to help generating more ideas, in the aggregate those ideas look alike:

First, having access to generative AI effectively equalizes the evaluations of stories, removing any disadvantage or advantage based on the writers’ inherent creativity. That generative AI particularly benefited less able writers is paralleled in recent studies focusing on other domains in which generative AI has been shown to help less productive workers. Second, one might ask whether the generative AI ideas can push the upper bound of creativity of produced stories, beyond what particularly creative humans are capable of on their own. We do not find evidence of this possibility in this study. […]

While these results point to an increase in individual creativity, there is risk of losing collective novelty. In general equilibrium, an interesting question is whether the stories enhanced and inspired by AI will be able to create sufficient variation in the outputs they lead to. Specifically, if the publishing (and self-publishing) industry were to embrace more generative AI-inspired stories, our findings suggest that the produced stories would become less unique in aggregate and more similar to each other. This downward spiral shows parallels to an emerging social dilemma: If individual writers find out that their generative AI-inspired writing is evaluated as more creative, they have an incentive to use generative AI more in the future, but by doing so, the collective novelty of stories may be reduced further. In short, our results suggest that despite the enhancement effect that generative AI had on individual creativity, there may be a cautionary note if generative AI were adopted more widely for creative tasks.

I wonder if two things are happening. One, the model is trained on something, it can only think with what it’s trained upon. Two, the optimisation pushes the model to converge on similar outcomes. Variety gets whittled out — unusual perspectives, diverse viewpoints, rare knowledge. It converges to the safe, statistical average outcome. AI researcher Andrew Peterson defines this as knowledge collapse.

Abstracting away from models, Lauren Leek recently wrote about what happens when prediction shapes the choices in front of us:

[V]ariety is where new options come from. The strange Georgian-Japanese street-food fusion restaurant nobody can classify, the pub that doesn’t fit the template, the song that sounds unlike what came before, most will never matter. But some will. A world that keeps removing the unlikely also removes the things that could have become the next likely. Ecologists have known this for a long time already and even priced it: they call it the insurance value of biodiversity, the standing reserve of rare species that cost energy and contribute nothing, until the climate shifts and one of them turns out to be what survives. A monoculture is efficient right up until the weather changes.

And there’s a second cost, harder to see, and it’s the one that actually a bit scared of. Once prediction shapes the choices in front of us, we lose the ability to tell the difference between what people wanted and what the system made easy to want.

Nobody starts in the right place. You don’t begin with the correct tool and work sensibly within its constraints until you organically graduate to a more capable one. That is not how obsession works. Obsession works by taking whatever is available and pressing on it until it either breaks or reveals something. The machine’s limits become a map of the territory. You learn what computing actually costs by paying too much of it on hardware that can barely afford it.

I know this because I was running Final Cut Pro X on a 2006 Core 2 Duo iMac with 3GB RAM and 120GB of spinning rust. I was nine. I had no business doing this. I did it every day after school until my parents made me go to bed.

I find this to be the defining difference between the “devices” (the phone, the tablet, the watch, etc.) and a computer: the computer is a blank slate, it is what you want it to be. It’s a portal to possibilities. The limit is how far you’re willing to tinker with it. The new devices — they’re more usable? Easier? Accessible to everyone? Harder to fuck up? — they have always felt optimised for consumption.

What if all of that immense bureaucracy, the endless processes, the mind-boggling range of costs that you need to make the computer compute, just goes poof? That doesn’t mean that the software will be good. But most software today is not good. It simply means that products could go to market very quickly.

And for lots of users, that’s going to be fine. People don’t judge A.I. code the same way they judge slop articles or glazed videos. They’re not looking for the human connection of art. They’re looking to achieve a goal. Code just has to work.

Andy Coenen:

Up to this point, writing software has been an arcane craft practiced by highly specialized artisans. Very few people can do it, and even fewer people can do it well. […]

Before the Industrial Revolution, the average person owned only a few pairs of clothes, and many people spent the majority of their lives making those clothes by hand. Today, very few people actually make clothes but there are thousands of apparel companies for every type of activity – from skiing to nursing to firefighting – that would have been unimaginable before. Similarly, the Software Industrial Revolution will lead to an explosion of new bespoke software across every industry and niche, and the fact that we’ll no longer build the software by hand means that we’ll build and use much more of it.

Matt Shumer:

I am no longer needed for the actual technical work of my job. I describe what I want built, in plain English, and it just… appears. Not a rough draft I need to fix. The finished thing. I tell the AI what I want, walk away from my computer for four hours, and come back to find the work done. Done well, done better than I would have done it myself, with no corrections needed. A couple of months ago, I was going back and forth with the AI, guiding it, making edits. Now I just describe the outcome and leave.

Nolan Lawson:

Someday years from now we will look back on the era when we were the last generation to code by hand. We’ll laugh and explain to our grandkids how silly it was that we typed out JavaScript syntax with our fingers. But secretly we’ll miss it.

Greg Knauss:

There are a thousand factors at play here (most of which are still in motion) but for plenty of small-scale, snap-together projects, something like Anthropic’s Claude Code or OpenAI’s Codex will be good enough, for economically-viable values of both “good” and “enough.” They’ll either burp up scripts that simply wouldn’t exist otherwise, or do (some of) the work of (some) junior or mid-level coders (somewhat) faster and cheaper. But the direction things are headed seems pretty clear.

Is the code any good? I don’t know. Who cares? Nobody looks at it anyway. AI produces a result, and results are what matter, and if you’re waiting for quality to factor significantly into that equation, I’ve got some bad news about the last 40 years of professional software development for you.

Tom Blomfield (Monzo’s founder):

I think we’re going to have a lot more software in the future. Normal, non-technical people will be able to summon up a new custom-written software program to solve a trivial daily task much the same way they might use a spreadsheet to make a list or do simple sums today.

Here’s the problem. That increased demand will be met with increased supply – from rapidly scaling AI. I just don’t think there will be any point in a human writing code for very much longer. My guess is that AI will soon be provably and obviously better at basically every facet of it. We will still have senior software engineers supervising the AI for many years because it makes humans feel safer. But they will eventually be like the conductors on self-driving trams; they are mainly there to make people feel better. This is the point that makes software engineers really angry, I think.

Steven Sinofsky:

There will be more software than ever before. This is not just because of AI coding or agents building products or whatever. It is because we are nowhere near meeting the demand for what software can do. This holds for software I use on my own, software a business needs, software an organization needs, or software to control the explosion of devices that replace every analog device with an automated one.

The animation was designed to mimic human breathing at 12 breaths per minute, and feel comforting and soothing.

I understand why it’s gone — our devices are always on, it’s not as much of a concern anymore — but it was the kind of nice little detail that set Apple apart, and made it more human. Everything looks so serious now.

Authority used to be the organizing principle of information, and thus the media. You earned attention by being right, by being first in discovery, or by being big enough to be the default. That world is gone. The new and current organizing principle of information is velocity.

What matters now is how fast something moves through the network: how quickly it is clicked, shared, quoted, replied to, remixed, and replaced. In a system tuned for speed, authority is ornamental. The network rewards motion first and judgment later, if ever. Perhaps that’s why you feel you can’t discern between truths, half-truths, and lies.

The network rewards reactions. Reacting precludes thinking: few people have the wit or knowledge to share something valuable on the spot, the rest sound like an indistinct, self-pleasing noise that contributes nothing.

This goes beyond media. People are getting sloppier at work, with Slack encouraging fast replies at any moment. They’re getting sloppier in their personal lives, where the expectation is to be reachable and ready at all times.

The network behaves like a gobbling No-Face because the network is designed to behave as such. All of the new popular social media platforms offer more of the same. What’s the point?