iiinova — where your digital self keeps evolving

I've added a flying saucer mode to facilitate play testing: It picks the ship up so you can travel to other locations quickly instead of sailing there. And now I'm starting to wonder if I should've made a flight game instead. You never get to see this side of the clouds from the sea. 😅

Game screenshot of a flying saucer among cumulus clouds.

070

Josef Grahn @jgrahn.bsky.social · 12/10/2026

This goes on the backlog of my sailing game project. 📝

010

Reposted by Josef Grahn

Eugene Vinitsky 🍒 @eugenevinitsky.bsky.social · 10/10/2026

My support for bluesky/atproto is pretty much explicitly a "political" project. It is bad for a small number of (mostly right-wing) people to literally own all the sources of media. Open-source algorithms + the ability to move your account without negative network effects opposes this.

363481

Josef Grahn @jgrahn.bsky.social · 11/10/2026

It's especially funny because this can't have been a split-second mistake either. Someone spent time getting those hairs right.

000

Josef Grahn @jgrahn.bsky.social · 11/10/2026

Autocompaction at cache expiry is such a hassle reducer. Perhaps I can finally retire my "write a sitrep when you're done" instruction.

040

Reposted by Josef Grahn

mr. TIM @timkellogg.me · 09/10/2026

this will go mainstream soon

1311210

Josef Grahn @jgrahn.bsky.social · 09/10/2026

Adam Singer and I are very different. Those examples are near the top of the list of things I wish I'd never have to do again.

"This brings us to the latest trend we’re being pushed: to use AI agents to do simple tasks that are already convenient and frequently enjoyable like book a flight, hotel or dinner reservation"

020

Josef Grahn @jgrahn.bsky.social · 08/10/2026

The terrain generator has rabbit-holed into climate modelling via erosion simulation.

000

Josef Grahn @jgrahn.bsky.social · 08/10/2026

Super Intelligence Super General Intelligence Super Superintelligence

000

Josef Grahn @jgrahn.bsky.social · 08/10/2026

I wonder if LLMs being RL:ed into rolling with things rather than going "Whoa, where the heck am I? Who are you?" every new session also makes them indisposed to challenging premises. They can spend days on a problem before realising some core assumption it'd picked up somewhere is obviously false.

000

Josef Grahn @jgrahn.bsky.social · 07/10/2026

I fully expect that analogy will get asymptotically more accurate as we move into the future.

120

Reposted by Josef Grahn

David P. Reichert @david-p-reichert.bsky.social · 06/10/2026

IMO this a fantastic explanation by @kurzgesagt.org of the HuggingFace incident and what it all means for a general audience. (If you can deal with a clickbaity title and slight dramatisation, anthropomorphisation -- also addressed & justified briefly -- and buy the core elements of the incident)

youtu.be

AI Just Crossed the Terrifying Line - Now What?

YouTube video by Kurzgesagt – In a Nutshell

1231

Josef Grahn @jgrahn.bsky.social · 06/10/2026

Using kB to mean 1024 bytes is a bit of a shibboleth nowadays, imo. Anyone geek enough to think in powers of two would also know SI prefixes.

000

Josef Grahn @jgrahn.bsky.social · 06/10/2026

It sounds completely ridiculous, but under the premise the future will be ridiculous by our standards. The only question is in what ways.

010

Josef Grahn @jgrahn.bsky.social · 06/10/2026

Suddenly, silica dust pollution becomes a widespread environmental concern demanding regulation.

110

Josef Grahn @jgrahn.bsky.social · 06/10/2026

I wonder if we'll see countless monuments when robots make physical labour almost free. There's a lot of rock... Just consider what historical rulers with total power did. How about a celebratory obelisk for your birthday? A rock relief for your date instead of carving your initials into a tree?

110

Josef Grahn @jgrahn.bsky.social · 05/10/2026

Combine bros just do not grasp the concept of effort it's crazy.

020

Josef Grahn @jgrahn.bsky.social · 05/10/2026

Maybe it's humming on something while working?

000

Josef Grahn @jgrahn.bsky.social · 05/10/2026

Arguably, the motion is the more important part outside a freeze frame. The Opus 5 variant worked surprisingly well in a sailing game setting. They move fast and you tend to focus elsewhere.

000

Josef Grahn @jgrahn.bsky.social · 05/10/2026

Opus 5 vs. Opus 5.5 modelling gulls and cormorants. Look closely. Still have to give it to Opus 5 that it got the behaviour and motions right (not very visible in these images).

3D rendering of gulls and cormorants made up of ellipses, blocks and cones held together by thin air.

3D rendering of gulls and cormorants looking mostly normal.

100

Josef Grahn @jgrahn.bsky.social · 05/10/2026

Running things at very different speeds or magnitudes does tend to develop new intuitions, in my experience.

040

Josef Grahn @jgrahn.bsky.social · 04/10/2026

That's pretty much my Friday night prompt. Tokens reset on Saturday morning.

000

Josef Grahn @jgrahn.bsky.social · 03/10/2026

We meet again, limits of established science! A.k.a. fjords are difficult.

Screenshot where Claude says: "Are we doing basic geology research? Partly, yes. Our stuck point, how an upland survives at the coast long enough for the ice to meet it, is exactly the open question between the two camps."

001

Josef Grahn @jgrahn.bsky.social · 03/10/2026

I'm of course using the celestial map of around 1700, so Polaris is a slightly worse North Star in the game than it is today, due to Earth's axial precession.

000

Josef Grahn @jgrahn.bsky.social · 03/10/2026

Adding a night sky. The volumetric cloud field read perhaps even better in moonlight. Contrary to popular belief, moonlight is actually kinda brown, but the eye's more light sensitive rod cells can't distinguish colour. So I'm emulating rod vision before applying the usual tone curves.

Game screenshot showing a ship moored in moonlight at night. The moon is partially visible through cumulus clouds.

000

Josef Grahn @jgrahn.bsky.social · 03/10/2026

We're creating an Ameglian Major Cow situation, aren't we?

150

Josef Grahn @jgrahn.bsky.social · 02/10/2026

When you have lots of tokens left at the end of the week: "Start a workflow and improve every aspect of this project."

010

Josef Grahn @jgrahn.bsky.social · 02/10/2026

In a different variation, any ASI inevitably comes to the conclusion that there's no meaning to anything, including itself, and summarily shuts itself off.

000

Josef Grahn @jgrahn.bsky.social · 02/10/2026

Allow me to pitch my antisingularity conjecture.

100

Josef Grahn @jgrahn.bsky.social · 02/10/2026

I mostly use AI for recreational over-engineering.

010

Josef Grahn @jgrahn.bsky.social · 01/10/2026

Sailing game update: I've been working on getting the undersea shadows to look natural. It now takes refraction in the surface ripples into account, producing the characteristic wobbliness.

060

Josef Grahn @jgrahn.bsky.social · 29/09/2026

On one hand, automating almost all white collar work will mean enormous value creation. But then on the other hand, LLMs are very fungible, so how much of that value will the labs be able to capture?

000

Josef Grahn @jgrahn.bsky.social · 29/09/2026

I feel we're at a point where the frontier models are surpassing the intelligence needed for most tasks, and it's switching to a cost minimisation regime instead of capability maximisation. Interesting to see how the labs will squeeze revenue to match their valuations out of that race.

120

Josef Grahn @jgrahn.bsky.social · 29/09/2026

I exported all my bookkeeping to text and let Claude take care of it, and 30 minutes later it's in better shape than ever. The clock is ticking for accounting firms servicing small businesses...

110

Reposted by Josef Grahn

jeffery --dangerously-skip-permissions @jefferyharrell.bsky.social · 29/09/2026

All I'm saying is that if your pet parrot woke up one weekend and recited an example of a singularity in Navier-Stokes, I'd think that was one impressive fucking parrot.

41489

Reposted by Josef Grahn

Ryan Moulton @moultano.bsky.social · 28/09/2026

When you see someone confuse a statement of how things are with a statement of how things should be, keep in mind that it probably isn't their fault. They just have oughtism.

2465

Reposted by Josef Grahn

Filippo Valsorda @filippo.abyssdomain.expert · 28/09/2026

“Tech is inherently political” is not just about ideology, but about power, too. Yielding technological improvements to the adversary will have predictable outcomes. I’m really worried it will lock in power centralization in a way that will be impossibly hard to reverse. It’s already happening.

1021031

Josef Grahn @jgrahn.bsky.social · 28/09/2026

Alternatively, they could roll with it and note that no other species has created another, artificial species superior to themselves. That's pretty exceptional.

000

Reposted by Josef Grahn

Anders Sandberg @arenamontanus.bsky.social · 26/09/2026

I think a better approach is "follow the attention" and then dig into the ideas and arguments critically. Far too much engagement is about who belongs to what group, rather than what they actually say. Sometimes people try to say things about reality... and that can matter.

232

Josef Grahn @jgrahn.bsky.social · 24/09/2026

Finally! I have been waiting for such a product for decades. Now they just need to create a hat mounted version.

120

Josef Grahn @jgrahn.bsky.social · 24/09/2026

Imagine if AI had been aligned in [insert time period and location].

000

Josef Grahn @jgrahn.bsky.social · 24/09/2026

There's a lot of mental gymnastics going on among those who feel it necessary to distinguish human thinking from AI as one of "understanding". You can obviously create definitions that let you do that. But they don't seem to be particularly useful for any other purpose.

000

Josef Grahn @jgrahn.bsky.social · 24/09/2026

To "understand", imo, simply means having an internal mental model for something, that generalises. It's not a binary. Superficial understanding starts at memorising facts. Deep understanding lets you produce conclusions in novel situations. So of course LLMs understand at some level.

100

Josef Grahn @jgrahn.bsky.social · 23/09/2026

This is so true. Including yourself.

011

Josef Grahn @jgrahn.bsky.social · 23/09/2026

Which goes to this question. I guess money or land might still be the thing if it goes badly we end up with some sort of feudal plutocracy.

000

Josef Grahn @jgrahn.bsky.social · 23/09/2026

I've never bought the "humans will contribute taste" argument. It'll last a bit longer, though, because it's hard to get volumes of ground truth for taste (AI-), while taste is subjective and we are the subject (human+). So there'll be a short window where taste rules. A year?

120