Compression made the information age possible by stripping things down to fit the pipes. Expansion made the AI age possible by blowing data back up again. Both operations leave marks; we’ve learned to spot compression artifacts, but we’ve only just begun to reckon with expansion artifacts. Until we do, there’s a lot of risk to manage.
People talk about the effectiveness (or lack thereof) of large language models as though all tasks are comparable. But it strikes me that there are three broad categories of work that large language models are applied to:
Compression is when you feed a large language model something big that you want to make small. Summarise this book. Give me the gist of this meeting. Large language models are generally pretty good at this, which makes sense given that they themselves are kind of like compressed artifacts.
Transformation is when large language models convert from one format into another. Turn this audio into text. Turn this jumble of data into structured JSON. A large language model can handle these tasks pretty well. There’ll probably be a few errors so make sure that’s not a deal-breaker.
Expansion is when you give a large language model a prompt to generate something from scratch. An image. A presentation. An email. A poem. This is where slop lives. The output inevitably betrays its origins, glistening with a sheen of mediocrity.
Laurie spotted this three-way split a while back:
Is what you’re doing taking a large amount of text and asking the LLM to convert it into a smaller amount of text? Then it’s probably going to be great at it. If you’re asking it to convert into a roughly equal amount of text it will be so-so. If you’re asking it to create more text than you gave it, forget about it.
I hope that when the bubble finally bursts, we’ll see the surviving large language models put to work on the first two categories. The boring stuff. The work that’s tedious for humans.
But tedious is as tedious does. Something I consider drudgery might be the very thing that gives you life. Like Giles says:
I have a feeling that everyone likes using AI tools to try doing someone else’s profession. They’re much less keen when someone else uses it for their profession.
The big exception seems to be programming. Apparently there are plenty of coders who never before expressed an interest in being managers who are now happily hanging up their coding spurs in favour being the overseer of non-human workers.
It’s a reasonable outlook. It could even be considered a user-centred approach. Users don’t care about the elegance of your code; they care about accomplishing their tasks.
Programming is something of an exception to the efficacy of large language models in general. Instead of relying on the subjectivity of painting, poetry, or prose, programming can be objectively tested. Throw enough money at the worst people in the world and they’ll give you tokens you can use to get the machines to test their own output. So you can get a large language model to create something reasonably good from scratch as long as that something is code.
If you had asked me about the threat model of large language models two years ago, I probably would’ve been worried for artists, writers, and musicians. I thought that software had enough inherent complexity to be relatively safe.
Now my opinion has completely reversed. Software is almost certainly the killer app for large language models.
I think the artists, writers, and musicians will be okay, or at least as okay as they ever were. It turns out that humans like things made by other humans.
And y’know what? If I had to choose which endeavour I’d rather see automated away—programming or art—it’s no competition.
Don’t get me wrong—it would be nice if everyone got paid for doing what they enjoy. It’s just that I’m okay with software engineers not being at the front of that line.
I remember when I first started getting paid money to make websites. “Really?” I thought, “Someone is willing to pay me to do something I’d do anyway?” I kept waiting for the jig to be up. Instead I saw my profession grow and expand.
Perhaps there’s a long-overdue compression happening.
Or maybe it’s more like a transformation.
Sometimes I look back through my blogging archives and notice what’s changed over time.
For example, I used to write quite enthusiastically about the arrival of a new operating system from Apple. That is no longer the case, to put it mildly. I’m currently holed up on Sequioa, trying to resist all the nudgings to “upgrade” to the tacky design nightmare that is Tahoe. I feel like the protagonist of Pluribus.
I used to write about software I really liked. Sometimes it was software made by Apple. More often it was from some independent developer.
Like, I remember how much I loved a little application called Quicksilver. It just did one thing. You pressed control and space and then started typing the name of any programme installed on your computer. After a few characters Quicksilver would show you the match, you hit enter and the programme launched.
If that process sounds familiar, it’s because Apple ended up incorporating it into their own Spotlight feature. Quicksilver got sherlocked (ask your parents).
Recently though, Spotlight got worse and worse at doing its one job. It’s been laggy and inaccurate, even though I set my Spotlight indexing options to only index the Applications folder.
Then I found TinyStart. It’s like Quicksilver reborn!
A tiny launcher for macOS, fast and focused on the essentials.
Actually, it does double duty. As well as being an application launcher, it’s also an emoji picker. 👍
Best of all, not only is TinyStart a return to the focus and quality of software of yore, it’s also a return to the pricing model. You buy the software—for a measly €5—and that’s it. You own it now. There’s no subscription you have to pay every month.
I love everything about this.
I feel very seen here. This describes how I built The Session:
There are still people building the web by hand, very much like we did it in the early days. They know all about what’s possible using modern tooling, yet they choose to expend their time and attention to the craft of doing it by hand. They care about the craft, and they care about what they’re making. They believe in their unique skill and vision over engagement strategies and analytics and content algorithms. They don’t need a platform, or they’ll build their own.
- I don’t and won’t use “AI” in the text of any of my published work.
- I’m not worried about “AI” replacing me as a novelist.
- People in general are burning out on “AI.”
- I’m supporting human artists, including as they relate to my own work.
- “AI” is Probably Sticking Around In Some Form.
- “AI” is a marketing term, not a technical one, and encompasses different technologies.
- There were and are ethical ways to have trained generative “AI” but because they weren’t done, the entire field is suspect.
- The various processes lumped into “AI” are likely to be integrated into programs and applications that are in business and creative workflows.
- It’s all right to be informed about the state of the art when it comes to “AI.”
- Some people are being made to use “AI” as a condition of their jobs. Maybe don’t give them too much shit for it.
Towards the end of 2025, I wrote:
I think I might change things up in 2026. Instead of waiting until the end of the year to write all the little reviews at once, I think I should write a review as soon as I finish a book. Instead of holding onto my reckons for months, I can just set them free one at a time.
I’ll get the ball rolling with the first book I read in 2026.
I’ve mentioned before that one interesting lens to apply to modern retellings of the Greek myths is how they treat deities. Are gods and goddesses real in this story? Or is it a non-interventionist tale with a purely human cast? In her book The Shadow Of Perseus, Claire Heywood wrote about Perseus, Medusa, and Andromeda without any supernatural characters. Having been impressed by that, I figured I’d go back to investigate her debut, Daughters Of Sparta.
The framing device is one I hadn’t come across before. It follows the diverging stories of sisters Helen and Clytemnestra, flipping back and forth between the two throughout their lives. I’ve read plenty of takes on the Trojan war, and I’ve read plenty of takes on Clytemnestra’s revenge, but I think this is the first time they’ve been combined like this.
Overall, it works. There are inevitable time jumps. Some time periods are bound to get more attention than others. And at some point, the narrative just has to wrap up, even though we know there’s pleny more that follows afterwards.
All in all, a good addition to the list of modern retellings of classical Greek stories.
Laura’s classic book is now a web book that you can read for free online.
The promise of AI – the promise AI companies make to investors – is that there will be AIs that can do your job, and when your boss fires you and replaces you with AI, he will keep half of your salary for himself, and give the other half to the AI company.
That’s it.
That’s the $13T growth story that MorganStanley is telling. It’s why big investors and institutionals are giving AI companies hundreds of billions of dollars. And because they are piling in, normies are also getting sucked in, risking their retirement savings and their family’s financial security.
Now, if AI could do your job, this would still be a problem. We’d have to figure out what to do with all these technologically unemployed people.
But AI can’t do your job. It can help you do your job, but that doesn’t mean it’s going to save anyone money.
I am interested in art—we are interested in art, in any and all of its forms—because humans made it. That’s the very thing that makes it interesting; the who, the how, and especially the why.
The existence of the work itself is only part of the point, and materializing an image out of thin air misses the point of art, in very much the same way that putting a football into a Waymo to drive it up and down the street for a few hours would be entirely missing the point of sports.
It’s wild what you can do with CSS these days!
The transcript of a very thoughtful talk by Frank.
Stick with this. It’s worth it.
A great interview with Ted Chiang:
Predicting the most likely next word is different from having correct information about the world, which is why LLMs are not a reliable way to get the answers to questions, and I don’t think there is good evidence to suggest that they will become reliable. Over the past couple of years, there have been some papers published suggesting that training LLMs on more data and throwing more processing power at the problem provides diminishing returns in terms of performance. They can get better at reproducing patterns found online, but they don’t become capable of actual reasoning; it seems that the problem is fundamental to their architecture. And you can bolt tools onto the side of an LLM, like giving it a calculator it can use when you ask it a math problem, or giving it access to a search engine when you want up-to-date information, but putting reliable tools under the control of an unreliable program is not enough to make the controlling program reliable. I think we will need a different approach if we want a truly reliable question answerer.
While I’ve been listening to Hounds Of Love, I’ve also been reading Orbital by Samantha Harvey.
Here’s a passage from an early chapter as the crew of the International Space Station watch a typhoon forming:
How wired and wakeful the earth seems suddenly. It’s not one of the regular typhoons that haphazardly assault these parts of the world, they agree. They can’t see it all, but it’s bigger than projections had previously thought, and moving faster. They send their images, the latitudes and longitudes. They are like fortune tellers, the crew. Fortune tellers who can see and tell the future but do nothing to change or stop it. Soon their orbit will descend away to the east and south and no matter how they crane their necks backward at the earth-viewing windows the typhoon will roll out of sight and their vigil will end and darkness will hit them at speed.
They have no power – they have only their cameras and a privileged anxious view of its building magnificence. They watch it come.
The penultimate track on Hounds Of Love is the magnificent Hello Earth with its eerie Georgian chant for a chorus, and magnificent uilleann piping from the late great Liam Óg O’Flynn on the bridge. It too features a narrator watching from space:
Watching storms
Start to form
Over America.
Can’t do anything.
Just watch them swing
With the wind
Out to sea.All you sailors, (“Get out of the waves! Get out of the water!”)
All life-savers, (“Get out of the waves! Get out of the water!”)
All you cruisers, (“Get out of the waves! Get out of the water!”)
All you fishermen,
Head for home.
Matching the song to the book feels like pairing a fine wine with a delicious morsel.
- Install Stuff Indiscriminately From npm
- Pick a Framework Before You Know You Need One
- Always, Always Require a Compilation Step
An enjoyable guided tour of album artwork starting at the beginning of the twentieth century.
We’re at a point in the most ecosystems where pulling in libraries is not just the default action, it’s seen positively: “Look how modular and composable my code is!” Actually, it might just be a symptom of never wanting to type out more than a few lines.
It always amazes me when people don’t view dependencies as liabilities. To me it feels like the coding equivalent of going to a loan shark. You are asking for technical debt.
There are entire companies who are making a living of supplying you with the tools needed to deal with your dependency mess. In the name of security, we’re pushed to having dependencies and keeping them up to date, despite most of those dependencies being the primary source of security problems.
But there is a simpler path. You write code yourself. Sure, it’s more work up front, but once it’s written, it’s done.
I like the look of this proposal that would allow authors to have more control over network priorities for third-party iframes—I’ve already documented how I had to use a third-party library to fix this problem on the Salter Cane site.
The tech bros advocating for generative AI to take over art are at the same level of cultural refinement as the characters in Severance. They’re creating apps to summarize books to people, tweeting from accounts with Greek statue profile pictures.
GenAI would automate Lumon’s cultural mission, allowing humans to sever themselves from the production of art and culture.
In which Rich nails Clearleft’s superpower:
“Clearleft is a relatively small team, but we can achieve big results because we are nimble and extremely experienced. As strategic design partners, we have a privileged position where we can work around a large company’s politics,” Rutter said. “We need to understand those politics — and help the client staff navigate them — but we don’t need to be bound by them. We bring a thoroughly user-centered approach to our design partnership, and that can be something novel to companies. By showing them what good design looks like (not so much the interface, as the actual process of getting to really well-designed products and services), we can be disruptive within the organization and leave them in a much better place.”