Your monthly non-spammy-gif-heavy newsletter from your friends at oio.

Hello oio friends! Welcome back to the thirty-fourth issue of oio news, the ever-growing dataset of first-class human slop from the depths of the interweb.

With all these GIFs, this newsletter is significantly larger than the 1997 PC game Need for Speed II. Since your browser has already downloaded it, you now have to read it to the end, unless you want to waste an all-time classic PC game.

It’s been a few months since our latest update, and a lot has happened. Another oio school wrapped up (yes, it was magical). We’ve been working alongside a company on a wild new music app, more workshops and public talks are piling up, and our little Roby’s growing up too.

But wait! We’re not going to talk about any of that today. That’s right, we’re saving these for the next newsletter, as something else has been quietly taking shape. Something we’ve been building in the background. Something special we really want to share…

… meet …

The Hot Air Factory!

The Hot Air Factory is our latest experiment! Have a look at the video here below, or just keep scrolling.

The cloudless cloud

Artificial intelligence surely feels like magic, but it’s not free. And we’re not talking money-free, but rather planet-entropy, energy free. Sure, we know it definitely heats up lots of air, but do you actually know exactly how much? Well, we didn’t. So we did what we typically do — we started prototyping!

If you’ve been following oio for a while, you’re well familiar with Roby, our AI creative director, forged in our Discord server, and recently moved from the cloud to a very real MacMini under our desk. Well Roby has recently been updated with an energy-tracking feature, where it tells you how much energy it used after each generation, in very simple human-readable terms: the amount of energy required, converted in espresso cups.

If you’re curious to look under the hood, have a look at the repo we cooked to make this work. Together with our creative technology Marta, we basically wrote a wrapper around MFLUX, with macmon tracking the energy.

but this was only the beginning…

The blinking box

The MacMini was just the first proof of concept, as we had one lying around in our studio, and it’s still powering Roby and all its little apps, but there are many baddies out there much more performant AND cheaper, and we can run all sorts of models on them.

In the end we went for an Nvidia Jetson Orin, the new family of boards, which for a few hundred bucks is an actual full-blown AI machine. Check out seeed studio, they sell a full line of fully assembled machines.

But you know we didn’t want just another blinking gray box… we’re a creative company on a mission to make technology less boring after all. So we started sketching.

In the end we explored cranks, dials, virtual grids, virtual factories, homes for Roby, it went WILD! But one day, looking at one of these car dealer inflatable guys, we had a spark—what if we could actually hook one of these up to our little server? wouldn’t it be funny? Oh wait! even better, what if we could hook up … a pinwheel?

and so it started…

The pinwheel

Remember Bjørn? He’s a design lead here at oio, and for a few weeks he became obsessed with pinwheels. Imagine mid summer July / August, asphalt was melting, most people were on holidays, not much work to do — the perfect time to embark on a side quest — design the perfect pinwheel. Something 3d printable but lightweight enough to spin, and most importantly that it would actually spin. And so it began. Together with our graphic designer Léane we all started brainstorming pinwheels, while 3D printing, sketching, modeling.

After many failed experiments, wheels not wheeling and pins not pinning, we finally got there: the perfect pinwheel ✨ 

The factory

So now we have this AI box puffing out hot air, but we needed something to tie everything together. A big, beautiful story, a metaphor. It was a fine line to walk, because we didn’t want to distract from the core issue: energy consumption.

We started hallucinating wild ideas, like a power grid simulator or a village of NPC peasants literally farming energy for the AI overlords. In the end, we settled on something simple, lightweight, airy, and cloudy.

But most of the credit goes to our superstar graphic designer Léane, who turned our messy thinking into a beautiful design, full of motion and illustration. Here some of the early tests, and check out the YouTube video below for the final thing.

But the pinwheel also needed a place to stay, and thanks to Bjørn again, we managed to create the perfect tiny factory to host our little AI engine:

How it works

The Hot Air Factory app connects to the AI computer inside the factory, and every time you prompt it it starts thinking. As the computer starts to think, it warms up and the internal cooling fan starts spinning. As the fan pulls cool air from the outside, the warm air is expelled through the top, making the pinwheel spin.

Energy consumption

The coolest feature of the factory is that it just shows you how much energy it consumes for every prompt, and it doesn’t share it in scientific measurements, but it converts it in everyday actions. For example:

A small image generation → ~0.06–0.07 kWh → An espresso cup

So you can now actually compare your daily AI usage with all your other daily actions, but most importantly you can now make informed decisions. It’s a way to make AI usage real and concrete by bringin it into the real world.

The Night Shift

One of our favorite features of the Hot Air Factory is the ability to run your prompts at night. For tasks that aren’t urgent or may require a lot of power, you can simply postpone them until nighttime, when the grid is less stressed and energy is cheaper. Just press the button in the app, and your prompt will bubble up into the ether. When night comes, the factory will light up and run it.

Is AI actually intelligent?

If you’ve been following big tech’s announcements about AI, you’ve probably heard something like: “this model has 7 billion parameters…” But if you’re human, that number likely means nothing at all.

In simple terms, parameters are what encode the model’s knowledge. The more parameters a model has, the longer it takes to train, and the larger and more computationally demanding it becomes. That’s why bigger models are associated with larger energy footprints. Not only during training and storage, but also during inference (when you ask them questions).

We thought it would be good for users to have a direct choice in which model to use in the Hot Air Factory. That’s why we picked three open-source models of different sizes and energy consumption. You can decide which one to run your prompts on — use less intelligence, save some energy, or turn it up, if you want a deeper thought.

The video, again

As promised at the beginning of this post, here’s the video again. If you watched it once, you can watch it twice. Maybe it’s not even the same video, is it? Only one way to find out.

How can i get one

Good question, you can’t. At least for now. We’ve only made a few, and the 3d design and BOM are still heavily work in progress, We don’t want to share something that may not work as expected yet. But we hope to put out something in the next few weeks / months! So be sure to join our discord and sign up to this newsletter.

And finally, why?

Another great question! And as design travelers, we all know too well that the answer to this question usually shows up late in the journey, in the dark woods, forcing design’s most sheepish soldiers to stumble through the forest of design iterations™. (this sentence makes no sense)

So the other side of the question: “why not?“ Why are we not doing something about this? Why are we just doom-prompting at every minor inconvenience in life?

We all use AI everyday, and we kind of know that it uses some energy, but we really have no way of knowing how much. And this is important, because without knowing we can’t make informed decisions, and environmental choices are actually important in these times we live in.

And you know, all of this adds up. In order to keep up with demand, we need more power. And to get more power, we need more data centers. And that’s how we got here. AI companies are running an arms race to build ever-larger models, which require more power, more electricity, more land. But the scale of these operations it’s nothing we’ve ever seen before. This is getting quite messy: grok’s data centers for example, had to run on methane, as they can’t get enough electricity. Fun! Paul Virilio, you would’ve loved this.

Cool! (hot?) now that the dystopia is out of the way — why are we actually doing these projects? Because that’s what we do here at oio. We make products with technology that embody a different vision, beyond the dominating capital-driven, efficience-ridden, speed-powered future visions. We make everyday products and tools for a less-boring future, and the Hot Air Factory is just one of many.

Take something that is far, obscure and incomprehensible and make it fun, local and accessible. That’s the oio’s playbook in a nutshell, and we are becoming quite good at it! Remember Updatables? (if you don’t, don’t bother clicking, SPACE10’s website links to a Peruvian betting company now (or lock in to some peruvian blackjack)).

If we did a good job, now you should feel more awake and aware about these issues. And if you actually read until here, it’s time to touch grass my friend.

oh wait, it’s time for some…

🍬 Cool links!

Your monthly supply of handpicked 𝕠𝕚𝕠 𝕔𝕖𝕣𝕥𝕚𝕗𝕚𝕖𝕕™ fine goods from our Discord community (not there yet? Join now 🏃‍♂️).

👨‍💻 dating according to browser history _{(social platform)}
🔠 funky letters _(www)
📥 a social network open only 3 hours per day _{(social platform)}
🐦 birds can save data _(YouTube)
👾 pioneers of computer graphics _(YouTube)
🖼️ reliving ancient times _(x)
🕹️ infinite mini games, vibe-coded _(x)
🪰 time flies, literally _(www)
🌈 queer AI _(www)
⛰️ NLP for indigenous language _(www)
☁️ the illusion of thinking _(apple.com)
👃 giving computers a sense of smell _(www)
📊 LLM running a Business _{(anthropic.com)}
🔗 LLM embeddings explained _{(huggingface.co)}
🚦 real-time interactive video _(www)
🎬 be a showrunner _(www)
📟 digital plastic _(www)
👁️‍🗨️ digital doppelgänger _(www)
🤳🏽 the year twitter died _(www)
🪶 how 2 poem _{(docs.google)}

Want to share your cool links? Join us on Discord!

📈 Monthly stonks

🌐 website: spiralgetty
🕹 game: admin
📁 arena: cubist google earth
📚 wiki: kamon
🔧 tool: packcad

A note on writing machines

I know the future will punish me for this (they already did) but I noticed how many people started equating the use of em dashes to the usage of LLMs, as these clankers tend to overuse them in their slop. So as you’ve probably seen — there are many em dashes in this newsletter, just like in all of oio’s previous newsletters, way before LLMs were a thing. Thing is — I love em dashes. but now that LLMs also started using ‘em recklessly, I don’t want you — our most valuable reader — to think that we’re using machines to write. We don’t. All of this garbage you are reading, is preem, uncut, often unedited, meatbag slop. So don’t get fooled by the em dashes, thanks for reading (hopefully human) friend. M

be cloud - M+N