Robin Sloan
the lab
December 2022

A year of
new avenues

A mostly realistic painting of a street scene in Paris, cobbles below, buildings in sharp perspective above, and right in front of us, a couple strolling, one of them holding an umbrella to ward off the rain, both looking off to the side, attention captured by something we cannot see. It is perfect flânerie.
Paris Street; Rainy Day, 1877, Gustave Caillebotte

It’s so power­fully obvious to me, it might as well be written in ten-foot letters of flame: the platforms of the last decade are done.

I said it in April 2022, and I believe it even more today: their only conclu­sion can be abandonment; an overdue MySpace-ification.

This is … tremendously exciting! Some of you reading this were users and/or devel­opers of the internet in the period from 2002 to perhaps 2012. For those of you who were not, I want to tell you that it was exciting and energizing, not because every­thing was great, but simply because anything was possible.

The concrete hadn’t set.

Now, after a decade of stuckness, the pavement is cracking — crumbling — and I want to insist to those of you who lived through that time, and those of you who didn’t: we all have a new opportunity.

In this newsletter, I want mainly to offer an exhortation. Then, because I can’t help myself, I’ll comment on an item of current interest. I’ll conclude with a technical question, because I know the right nerds are reading.

This message was emailed to lab newsletter subscribers. The assumed audience is subscribers who feel invested in the future of the internet who also have the capacity, and/or curiosity, to tinker on new projects. (Here’s more about assumed audiences.)

Here’s my exhortation:

Let 2023 be a year of exper­i­men­ta­tion and invention!

Let it come from the edges, the margins, the provinces, the marshes!


I am thinking specif­i­cally of exper­i­men­ta­tion around “ways of relating online”. I’ve used that phrase before, and I acknowl­edge it might be a bit obscure … but, for me, it captures the rich overlap of publishing and networking, media and conviviality. It’s this domain that was so deci­sively captured in the 2010s, and it’s this domain that is newly up for grabs.

It is 2003 again. Facebook, Twitter, and Instagram haven’t been invented yet … except, it’s also 2023, and they have, so you can learn from their rise and ruin.

This doesn’t mean you ought to start a company.

As the platforms of the last decade crumble, we might put “founder” culture back on the shelf. Startup finance works fine for building a business of a very particular kind; and, like, thank you for Shopify! Seriously! But, for a decade, this very partic­ular kind of business had a lock not only on internet commerce, but internet culture, too, with only ill effect.

I want to insist on an amateur internet; a garage internet; a public library internet; a kitchen table internet. At last, in 2023, I want to tell the tech CEOs and venture capitalists: pipe down. Buzz off. Go fave each other’s tweets.

It’s plain that neither the big tech companies nor the startup financiers are going to produce the “ways of relating” that will matter in the next decade. Almost by defi­n­i­tion, any exper­i­ment that’s truly path­breaking and provoca­tive is too weird and tiny for them to suffer. They are trapped in their stupen­dous scale; lucky us.

We should say: thanks for the browsers. Thanks for the seamless global slabs of compute, available for pennies and less. We’ll use them.

A whale dies in the ocean, and its carcass feeds a whole ecosystem for decades.

The browsers: let’s dwell there for a moment. Because “browsers” doesn’t just mean Chrome, Safari, and Edge; every operating system now offers a trivially available web canvas to which you can add anything you want. Aldus Manutius, who printed the first pocket-sized books in Venice at the turn of the 16th century, would have wept to possess a “page” of this kind.

Next, he would have crit­i­cized it. But first, the weeping!

All along, from the frothy 1990s to the perco­lating 2000s to the frozen 2010s to today, the web has been the sure thing. All along, it’s been growing and maturing, sprouting new capabilities. From my vantage point, that growth has seemed to accel­erate in the past five years; CSS, in partic­ular, has become incred­ibly flexible and expressive. Maybe even a bit overstuffed — but I’ll take it.

For people who care about creating worlds together, rather than getting rich, the web is the past and the web is the future. What incred­ible luck, that this open, decen­tral­ized “way of relating” claimed a position at the heart of the internet, and stuck fast. The web is limited, of course; frustrating; sometimes maddening. But that’s every creative medium. That’s life.

Okay, so, what kind of exploration? What kind of invention? I have some rough (and perhaps only margin­ally useful) notions for new avenues, which I will type out below.

First, though, I want to acknowl­edge that inventing new things, partic­ularly new “ways of relating”, is a lonely task. In the beginning, it’s just you, or, at best, you and your tiny gang of collab­o­ra­tors. You can only HOPE that perhaps a few dozen other people start to pay attention. That doesn’t feel like anything real. It’s difficult.

And … that’s just the way it goes!

The good stuff is always lonely in the beginning. Nothing mean­ingful will get made if its potential makers all wimp out too soon, because they get scared by the sparse crowd, the empty room.

You have to take on the loneliness. You have to strengthen that muscle.

Maybe I can help. If I am so hotly demanding a year of new avenues, the very least I can do is keep count. In 2023, if you make something, or even write about making something — if you take the time to describe your hopes and desires for the internet — please send me a link. I’ll circulate it in an edition of this newsletter, which will, throughout 2023, go out on the last Saturday of every month. Maybe some months there won’t be much to share; that’s fine. I’ll keep sending.

As I’m writing this, about 4000 people are subscribed to this lab newsletter. That’s not a lot, but maybe it’s the right 4000 people.

Here are those avenues.

A realistic painting of several young men, shirtless, planing a wood-planked floor smooth. It looks like difficult work!
The Floor Scrapers, 1875, Gustave Caillebotte

Try the new new new thing

Spend some time with Arc, the new browser from The Browser Company of New York. It’s an opin­ion­ated appli­ca­tion that’s constantly flexing and morphing as the team embroi­ders fresh ideas. Using the app at this stage in its devel­op­ment feels almost like following the new season of a TV show.

The original web browsers, back in the 1990s, suggested new activities, new problems, new opportunities. Does this next-generation web browser suggest anything new? I’ve only used it for a little while, and already I think the answer has to be “yes”. (Don’t miss the uncanny cutouts of live web pages available in Arc’s “easel” function … )

Arc is currently available only for Mac, and there is a waitlist. (If you received this newsletter via email, go check the copy in your inbox for an invite link.)

Think deeply about discovery

How might you help people find new things on the internet? How might you give new things on the internet a mean­ingful audition, without turning it all into a game that can (and will) be hacked and mastered?

I suspect the best answers are grounded in good old-fashioned human recommendations, but who knows? Maybe TikTok has it right; maybe everybody deserves one (1) audition with the capri­cious god-algorithm of the realm. Maybe there’s something to be isolated and improved there.

Either way, this is a big deal. Publishing on the internet is a solved problem; finding each other on the internet, in a way that’s healthy and sustainable … that’s the piece that has never quite fallen into place.

Climb into an overlay

The deep structure of the internet stymies peer-to-peer protocols; I wrote about this earlier in 2022.

In response, some people are turning to simple, secure overlay networks that make peer-to-peer addressing easier, or indeed, even possible at all. ZeroTier and Tailscale are the leaders here, and they’re both very impres­sive services. A few years back, I set up a ZeroTier overlay network just for my own computers (laptop, desktop, Linux box, cloud instance) and it has been a tiny revo­lu­tion in my technical work.

I am 100% convinced there are inter­esting new “ways of relating” waiting on these overlay networks, these pocket internets. They might require config­u­ra­tions beyond what’s currently supported, but these companies are suffi­ciently young and flexible that, presented with a compelling idea, I get the sense they’d just … implement it!

Maybe you should set up a ZeroTier or Tailscale network with a few friends or collab­o­ra­tors and … see what happens?

Go digging in the crates

What half-forgotten old protocol might be revived and repurposed?

Consider the example of RSS. Designed for blogs and news websites, its greatest success had nothing to do with those formats: RSS is the protocol that powers podcasting, all of it, under the hood, invisibly.

Where else is the wiring already in the walls? What new signals might you send?

My own avenue of exploration, a new protocol called Spring ’83, takes direct inspi­ra­tion from the internet of forty years ago, in partic­ular the hilarious simplicity of RFC 865, the Quote of the Day Protocol. I think there’s a lot of energy still coiled up in those early RFCs.

Make a thing with which you can talk about the thing

I feel like this is a common pattern: a community is building something new, and they talk about it … on Twitter. Maybe Discord. That’s fine, obviously, but there’s a sense in which it gives away the game on the first move.

Back in the 2000s, a lot of blogs were about blogs, about blogging. If that sounds exhaust­ingly meta, well, yes — but it was also SUPER generative. When the thing can describe itself, when it becomes the natural place to discuss and debate itself, I am telling you: some flywheel gets spinning, and powerful things start to happen.

This is related to my opinion that the very best movies are about movies, the very best books about books.

When the meta conver­sa­tion happens elsewhere, it’s like a little energy leak. Maybe that’s the energy you need.

Make exemplars before services

Remember that a single exemplar of a new format can be a profound contribution; in art and culture, maybe the MOST profound. These are the works that found genres.

Blogs are a useful example. The first blogs were static web pages, edited by hand, with new posts appended to the top. It was only after the format was proven out that other people — different people — came along and built software and hosting services expressly for blogging.

If you want, you can be that first example! You can edit by hand!

Work with the garage door open

Less an avenue, more a way of approaching an avenue:

This isn’t a time for “products”, or product launches. It’s not a time to toil in secret for a year and then reveal what you’d made with a shiny landing page.

Rather, I believe it’s a time to explain as you go. Our “work”, in an important sense, is to get into each other’s heads; to blast out cosmic rays that might give rise, in other minds, to new ideas.

It’s been a decade of products, smooth and sleek; apps with chamfered edges. I am inter­ested now in visions, compulsions, provocations.

Patrick Tanguay on Umberto Eco:

[Eco] uses Dumas’ The Count of Monte Cristo, Dante’s Divine Comedy, Shakespeare’s Hamlet, and Curtiz’s Casablanca to show that cult classics are cults “precisely because they are basically ramshackle, or ‘unhinged,’ so to speak.” It’s their imperfectness, the disjointed parts, that gives fans something to attach to, something to remember, something to cite.

I am inter­ested in something to cite!

Don't settle for Mastodon

I suppose this is an anti-avenue, because: Mastodon is not it.

When you tell me about Twitter vs. Mastodon, I hear that you got rid of the flesh-eating piranhas and replaced them with federated flesh-eating piranhas. No thanks, I’m still not swimming in that pool!

I’m not saying you shouldn’t create a Mastodon account, or that you can’t enjoy fun, perco­lating conver­sa­tions on that platform. I’m just saying that it does not, to me, represent a suffi­ciently inter­esting exper­i­ment, because it accepts too much as settled.

The timeline isn’t settled.

The @-mention isn’t settled.

Nothing is settled. It’s 2003 again!

That’s it for my avenues. Again, I don’t have any great confi­dence they will be inter­esting or moti­vating to anyone else; these are just the thoughts and intu­itions to which I’ve found myself returning, over the past year or so.

Now, a comment on the fizzing thing of the moment!

Automation in the real world

An impressionistic painting of a man walking in the streets of Paris. The painting is very accomplished, but it has the tell-tale gooiness of AI art.
The prompt "an impressionist painting of an elegant man strolling the streets of Paris, at night time, under warm street lamps, with wet cobblestones on the street, by Gustave Caillebotte" rendered by Stable Diffusion 1.4; author's favorite from ~400 variations, of which many were terrible

Quickly, a thought related to ChatGPT, OpenAI’s new language model splashily released for all to try.

It’s an impres­sive system — the brightest premo­ni­tion yet of something that will be part of human life, for better and for worse, from here on out.

But I guess I’ve been using these AI tools long enough that the “wow” has worn off, so I’m left with the “what now?”

Generative models of this kind produce results that can be startling, spooky, lovely, or all three. I have a directory bursting with images created using Stable Diffusion. They were fun to make, but/and … what are they for? So far, the answer is “my amusement”. That’s fine; I like being amused. But amusement is not what’s stenciled over the door at OpenAI. They want to remake the whole economy over there.

ChatGPT’s release has triggered gleeful pronounce­ments that whole cate­gories of human activity will soon (as soon as this weekend?) be made obsolete. I want to suggest that this isn’t how automa­tion works; in fact, it ignores every­thing about the process that’s actually inter­esting.

Take the example of container­ized shipping. This is textbook automa­tion, but the picture is not “replacement”, a line of clanking robo-longshoremen carrying bulky cargo from the bellies of ships. Rather, an entirely new task was imagined — moving and stacking enormous, modular containers — and it’s THAT task that was given to machines. This is hilarious, if you think about it: “Ah yes, we automated this onerous human labor … by completely rebuilding the entire physical infra­struc­ture of global shipping, fighting a pitched political battle along the way!”

(Today, certain container ports are among the most automated spaces on earth, with bright lines painted on the ground indi­cating zones of machine activity. If you step across, there’s a good chance you’ll get squished.)

Likewise, here’s a lesson from my work making olive oil. In most places, machines have taken over the olive harvest, but that’s only possible because olive groves have been planted specif­i­cally to fit their shape. A grove planted for machine harvesting looks nothing like a grove planted for human harvesting. Again, the hilarity: “Tired of picking those olives by hand? These mighty harvesting machines are here to take over — and they’ll do it ten times as fast! Now, if you’ll just rip all your trees out of the ground … ”

One more example, courtesy of Makshya Tolbert: the cotton gin, a canonical example of automa­tion, perversely increased the demand for slave labor because it made cotton growing so much more profitable.

So, the obsolence of a partic­ular human task is, like, the LEAST inter­esting thing about these processes, and cackling about the end of task X or job Y is like staring at a spot on the carpet while they rebuild the whole house around you. When you encounter someone doing that, I encourage you to ignore them, because it’s a sign they’re not a serious thinker, and/or they’re not curious about how the world really changes.

For me, the inter­esting questions sound more like

That last question will, on the timescale of decades, turn out to be the most consequential, by far. Think of container ports, or olive groves, or better yet, think of cars, and how dutifully humans have engi­neered a world just for them, to our own great detriment. What will be the equivalent, for AI, of the gas station, the six-lane highway, the parking lot?

I have to confess: even consid­ering ChatGPT’s prodi­gious capabilities, I’m skeptical that any of this will happen very quickly. That’s because OpenAI’s GPT-3 has been around for two years now. It was, and is, extremely capable … yet I’m not sure what it has accom­plished in the world. And even if ChatGPT and the forth­coming GPT-4 are expo­nen­tially more capable, you do need some base value to, uh, exponentiate.

Maybe some number of companies have put GPT-3 to work behind the scenes, spewing out language, hugely profitable, entirely secret. I don’t think that’s the case, but of course it’s possible. I would love to know!

It is telling, though, that its public appli­ca­tions all have to do with the mass produc­tion of medium-quality text … which wasn’t exactly rich terrain even before GPT-3 came on the scene. The GPT-3-powered story­telling game called AI Dungeon was briefly dazzling, but it seems now to have faded away.

What can these systems do, after all the screen­shots have been taken?

The milestone that will impress me, honestly, is an AI agent with a bank account. Obviously that will also terrify me. But/and, maybe this is Sloan’s defi­n­i­tion of arti­fi­cial general intelligence: given a bank account and a few bucks to start, can this thing make it in the world?

That’s the task we’re all given, isn’t it?

For me, ChatGPT’s fluency with code is its most impres­sive feature. If you spend any time with it, try asking politely for “some cool CSS code for a blog”, or “a JavaScript function to identify an email address in a string”. It’s really something! Imagine an AI agent with internet access, a bank account, this fluency, and, crucially, some way to make a plan, to structure and sequence its actions … 

That leads us pretty quickly to the vision that Jack Clark conjures in a striking micro-story appended to the latest edition of his AI newsletter. The story concludes:

When I logged out, I went back to the regular internet. Since the AI models had got minia­tur­ized and prolif­er­ated a decade ago, the internet had radically changed. For one thing, it was so much faster now. It was also dangerous in ways it hadn’t been before — Attention Harvesters were every­where and the only reason I was confident in my browsing was I’d paid for a few protection programs.

Automation isn’t really about the end of task X or job Y; it’s about remaking the world to fit a new machine, for better and for worse.

December 2022, Oakland