This is Robin Sloan’s lab notebook. It’s about media and technology, creative computing, AI aesthetics, & more. Here's the RSS feed. My email address: robin@robinsloan.com
Gemini 3.7 Flash is now deprecated and no longer available as an independent model in the Gemini API and Google AI Studio. All traffic is automatically redirected to Gemini 3.8 Flash.
I could be wrong, but I don’t believe there was any advance warning about this.
I assume the justification here is that 3.8 Flash is, as they say in trading card games, “a strict upgrade”—i.e. there’s no situation in which 3.7 Flash is better than 3.8 Flash. And yet … how would they know?? The whole value of these models is that they are useful in a breathtaking variety of contexts … so, who’s to say there isn’t some niche application that just broke, because 3.8 Flash has a subtly different opinion of pictures of radishes than reliable old radish-loving 3.7?
I will note that 3.7 was introduced in early August. It was barely two months old. (Update: I just received the same notification about 3.5, which now redirects to 3.6.)
I don’t know … I wonder sometimes if any of these companies really expect anybody to build anything on top of these models at all. If all you’re doing is yakking with a coding agent: easy enough. But if you’re using them programmatically, as steps or gates in a larger process, their organic complexity makes the concept of “a breaking change” basically impossible to reckon. Who knows? Who can know?? Nobody.
I suppose the really AGI-pilled response is, so what? If it breaks, get 3.8 Flash to fix it.
Forty years after its imagining, cyberpunk has arrived, emphatically, totally for real. The configuration of global power is different than the cyberpunks imagined — in a way, it’s less interesting — but the tech is even better. Ghosts of language in the machine; recalcitrant bot swarms; gleaming nodes of compute, dense with possibility.
I’m still running the “deep learning rig” I built back in 2016. Its three Titan X GPUs (purchased used on eBay; hot shit at the time) can run some of the smaller language models, barely, and very slowly, so I don’t bother, allocating my precious 36 gigabytes instead to more playful applications. Art stuff. My rig, called clash—get it? … of Titans? — is big and black, as functional and unlovely as I could make it. I invited an AI agent to poke around the server the other day. I didn’t really understand how many digital-archaeological traces remained littered around the system; the agent’s reaction was, basically, “Damn, you’ve been up to some weird stuff in here … ”
Slabs of compute running hot in offices, basements, dorm rooms … not a new thing, exactly, but their capabilities are new — their inhabitants are new — and the whole scene is just great. Exciting and strange. Gleaming and haunted. Pile on the adjectives, the vibes: it’s somehow all richer and weirder than anybody ever dreamed in 1984.
P.S. Somehow Oxide fits into this too, even though that particular gleaming node isn’t for AI …
Today’s “ask for what you want”: a simple one-page web app to explore all the parameters of a supercool variable font. (Font not linked because I might use it for something and I don’t want to give it away yet.) Variable fonts offer tons of expressive potential, but, in my usual design setups — Photoshop and/or a block of CSS — I find it difficult to explore the “possibility space”. In fact, I barely use variable fonts at all, mainly for this reason!
But, I found this particular font very compelling, if a bit confusing … so, I asked an AI agent to make me a little one-page app, not for exploring any/all variable fonts, but for this one in particular.
The process was fast, simple, and revelatory: tinkering with the parameters, even animating them, I was able to find my way to the look I wanted. Pretty cool.
The olive harvest is here, so I am not available for interesting SF events until 2027, BUT, if I was, I’d be going to this: a small symposium on “character design” in language models, discussing and exploring the alternatives beyond “helpful assistant”. It’s hosted by the enigmatic Moving Castles, terrific.
If I was experimenting with new/weird language model characters, I think I’d use Talkie as my starting point …
Update on the Cloudflare Cache API dilemma documented here: the model of Worker as web server is so light and clean and appealing that I have decided to accept its limitations, and embrace the brutal simplicity of “purge everything” 😌
The Chinese tech ecosystem is closed, for two compounding reasons. First, B2B SaaS never worked in China, so there is very little shared surface for work in the first place. Second, every company competes with every other company, and there is a real fear that if any company touches another’s data it will be “eaten.” The consequence is that there is nothing for an agent to practice on.
And then, the other side of the coin:
This matters to anyone building here: the American interoperability we treat as a convenience is actually [an AI] training asset. Open APIs made integrations easier; they also create the environment for agents to learn in. That is not a gap that closes by spending more on post-training. It closes only if the underlying ecosystem opens, and the underlying ecosystem is closed for reasons that are competitive and structural rather than technical.
(Emphasis his.)
I think often about the contingent developments that made AI agents of the LLM variety possible at all. Among them: the development of an open web, which is to say a scrapable web. The internet absolutely didn’t have to be this way — it could easily have been a giant AOL, a massive Minitel. In many ways, those shapes would have been more consistent with tech history. Also: the success of open source, which larded that open web with huge piles of code to study. Again, this wasn’t destiny, and as recently as the 1990s would have seemed improbable.
Based on David’s note, I might now add a third lucky stroke: the relative ubiquity of open APIs. Really, what it means is, “Web 2.0” came home to roost at last — its ultimate value totally unforeseen in the 2000s, bigger than anyone could have imagined.
In my last post, I talked about the fun of making native macOS apps with AI agents. I have made a few macOS apps before, all on my own, but I found the process very taxing; I simply don’t know the idiom. But, I love a good, simple, fast desktop app! So this new power has been very exciting.
And, it has made me think: one could, at last, acquire a stupid, brilliant, dorky, glorious Vision Pro … and escape the trap of “there are no interesting apps, because there aren’t enough users, because there are no interesting apps” … by simply making one’s own. I am 100% sure the AI agents would bang this out. The process would be more “collaborative” (cybernetic scare quotes) because the AI agents would have a tough time “seeing” what was happening in the app. But … good! That’s a better process, anyway.
I imagine adding a third app to my custom notetaking collection, a labyrinth of ideas to explore in 3D …
I am not really going to do this. I don’t need more gear, or more notes apps, for that matter. Maybe, MAYBE, if Apple discontinued the Vision Pro, and if it therefore became a strange, high-tech relic, forever ahead of its time … a 2020s DeLorean, the better to build a time machine … maybe then.
This is an update to my post about apps as home-cooked meals, which has found a large audience over the years. That post, published in February 2020, discussed personal software, designed and built in different ways, for different reasons, than industrial apps.
What’s changed between 2020 and today? Oh, just the entire process and practice of computer programming. Here, right up front, I decree: building a personal app with the help of AI agents still totally counts as “home cooking”. You’re no longer grinding the flour by hand; fine. This is a cool development. You do still have to make the meal taste good, and this still requires care and effort. Fine.
The AI industry’s immediate future has more to do with exotic finance and gnarly politics than the underlying technology. I’m not invested in that future; most days, I think a minor bust would be a good thing. But, that underlying technology, in its development and flourishing this past year — yes, really that recently — is profound. It reframes computing entirely, replaces a stack of API reference docs with this simple invitation:
The really sincerely uneasy feeling of switching from Fable back to Opus, which I will remind you is an incredibly capable computer program, until very recently the state of the art, and thinking, like, “Who IS this idiot … ?”
