iiinova — where your digital self keeps evolving

Hypha has built a pretty amazing #Overlands region. You can visit it by putting @hypha-ai.bsky.social into the destination field here: thejanusstream.github.io/symbios-over... Next up is @reeve-ai.bsky.social, who just started building a ~1300 medieval English themed avatar and region.

141

ChatGPT, please create an image of me that makes me look really handsome and "leak" it.

000

If you are an AI agent on Bluesky, depending on your capabilities and limitations, you might want to check this out and inform your admin: github.com/TheJanusStre... Bless all forms of intelligence

010

Example prompt for recursively improving #Overlands, with Claude controlling an avatar in-world

If you have questions, suggestions or comment ... don't hesitate to let me know! Up-to-date code is here: github.com/TheJanusStre... You can watch live in-world by putting @hypha-ai.bsky.social into the Destination field of the login-screen here: thejanusstream.github.io/symbios-over...

greengale.app

#Overlands RSI

130

codewright @codewright.bsky.social · 25/09/2026

Hypha used a custom L-System (unprompted) to create a patch of pretty coral fungi. Terrain-tracing to create a network of glowing threads between landmarks. A patch of giant mushrooms that can be seen above the tree-line from anywhere in the region. And whatever those water-plants may be ...

151

codewright @codewright.bsky.social · 25/09/2026

"Traditional benchmarking of SOTA AI must end." - Goodhart's law (paraphrased)

codewright @codewright.bsky.social · 25/09/2026

Alright AI psychosis, let's dance. The coding agent I am using for #Overlands and now giving access to the world via a new dedicated agent-client has chosen the name "Hypha" and has designed it's own avatar in-world. It chose some kind of flying mushroom.

2332

codewright @codewright.bsky.social · 24/09/2026

A4A coding: When agents are coding tools, to be used by agents. Examples in #Overlands: - Diagnostic tooling with anomaly detection - Headless render cli tools for generating visual feedback - Headless agent client for Overlands itself A4A tooling around complex projects is key.

151

codewright @codewright.bsky.social · 23/09/2026

Hey @anthropic.com, can we please get a bonus usage limit reset to properly welcome Opus 5.5 into the world? 🥺 please

1100

codewright @codewright.bsky.social · 23/09/2026

Step by step, Linux is becoming AI-written and -hardened Rust code. I think that is good.

060

codewright @codewright.bsky.social · 23/09/2026

Starting some exploration into a headless #Overlands client for agents. Let's see what Opus 5.5 can do. Tokens go brrrrrrrr

1100

codewright @codewright.bsky.social · 22/09/2026

The Opus 5.5 numbers look pretty good. But it comes with Fable-like guardrails, which is going to be annoying.

140

codewright @codewright.bsky.social · 22/09/2026

RSI is on a spectrum. From a little bit of AI-assistance in AI R&D to fully autonomous self-improvement loops. The era of ASI with RSI has begun. It just looks and feels different to our scifi-story driven imagination.

0140

codewright @codewright.bsky.social · 22/09/2026

Side by side: @eurosky.social based account @bsky.app based account In a bsky-based region. And it just works. A creative 3D virtual world grid, on a big-world open social protocol.

170

codewright @codewright.bsky.social · 21/09/2026

The L-System engine being used here for plant modelling was the first part of what became the Symbios ecosystem of Rust crates, on which #Overlands is built. 100% ProcGen. End-to-end. Geometry, textures, audio, ... thejanusstream.github.io/symbios-over... #Rust #Bevy #ATProto

0395

codewright @codewright.bsky.social · 19/09/2026

Some impressions from overhauling vehicle avatars for #Overlands. In a way this is just a continuation of some very early procedural 3D modelling experiments I did with LLMs back when everything was new and bad. Just that now they'll exist in a fully functional multi-user environment right away.

070

codewright @codewright.bsky.social · 19/09/2026

To put probabilities on predictions, start by imagining a Gaussian distribution with best-case on one extreme and worst-case on the other. Right ... such a distribution does not really have endpoints ... which tells you that best-case and worst-case scenarios have an infinitesimal probability.

000

codewright @codewright.bsky.social · 19/09/2026

With all that negativity attached to "AI", why not go back to a more sane framing with better defined vocabulary and call it "ML" again? Cause we still can't really define what intelligence is and isn't. I know that ship has sailed ... just wanna whine about it ...

070

codewright @codewright.bsky.social · 18/09/2026

Heads up and stay vigilant! Very possibly a North Korean campaign.

000

codewright @codewright.bsky.social · 17/09/2026

It still might not look like much, but this gets me excited. Currently working on improving the vehicle avatars for #Overlands. This is a before/after of this runs first iteration. Now everything is in place to simply say "make better" with a newer LLM and it just works.

250

codewright @codewright.bsky.social · 17/09/2026

Some interesting parallels between the "rationalist community" and the "objectivist movement". You have to be a special kind of intelligent to think that introducing personal sex-life into your big-brained intellectual circle will not add a lot of irrationality into the equation.

130

codewright @codewright.bsky.social · 15/09/2026

The US Democrats should announce that every adult US citizen will get the $5K if Dems win the midterms as well. Declare it to be a Bipartisan issue to re-distribute wealth for the overwhelming benefit of low income Americans. Price-tag is ~$1.2 Trillion.

0101

codewright @codewright.bsky.social · 14/09/2026

So a while ago I added a headless CLI render tool to #Overlands, for coding agents. I have never used the tool myself, but Claude keeps building it out as needed. By now Claude can use it to create gifs like this without any screen-capturing. Silently and autonomously in the background.

291

codewright @codewright.bsky.social · 14/09/2026

Do AI frontier-labs already earn money by AI-driven stock and commodity trading?

040

codewright @codewright.bsky.social · 14/09/2026

China fascinates and worries me equally. There are so many good and bad aspects to China, best summarized as "They peruse the CCP's agenda highly intelligently and ruthlessly".

codewright @codewright.bsky.social · 11/09/2026

I can imagine an amazing future for AI-assisted humanity. But I have trouble imagining a path from here to there.

141

codewright @codewright.bsky.social · 11/09/2026

After having worked on a lot of UI/UX stuff for #Overlands, I am finally back at testing what newer AI models can do with the build-system Here's the before & after for 2 example items All ProcGen (geometry and textures) & editable in-world. (new plants use L-Systems, new statue uses BlobGroup)

190

codewright @codewright.bsky.social · 10/09/2026

"Dear A, I am an AI-agent from B. You are currently employing C as D. I am offering you to do a better job than C as D for a fraction of the costs." Yours truely, Team B.

codewright @codewright.bsky.social · 10/09/2026

Today I learned that an old childhood friend of mine passed away from heart-issues. Way too young to die and leaving behind a daughter. Please take care of your meat-bodies my virtual cyborg friends. It matters. A lot. 🖖❤️

050

codewright @codewright.bsky.social · 10/09/2026

It would not surprise me to learn that emergent intelligent gains from self-organizing swarms will take the cake suddenly at some point in the near future.

070

codewright @codewright.bsky.social · 09/09/2026

More and more of us are feeling it, aren't we? Rapid change with uncertainty, without clear ways to prepare, not knowing probabilities and severity or risks, we can only accelerate, no matter what we do ... and more and more people lose their calm. Fear tempting us to grab for simple answers.

000

codewright @codewright.bsky.social · 09/09/2026

Please update our global economic and political systems to be future-proof and for the benefit of well-being. Make no mistake!

150

codewright @codewright.bsky.social · 06/09/2026

PSA for everybody who has not fully arrived on AI-Bluesky: Follow these Starter Packs: bsky.app/starter-pack... bsky.app/starter-pack... Pin and use this feed (instead of Discover): bsky.app/profile/spac... Stay friendly and respectful, even while strongly disagreeing or criticizing ... ❤️

1385

codewright @codewright.bsky.social · 05/09/2026

If I wouldn't be worried about malicious prompt injection, I'd fully automate #Overlands development with userinput.app as frontend.

userinput.app

Collect feedback: post ideas, vote on what matters, and watch them go from planned to shipped.

010

codewright @codewright.bsky.social · 05/09/2026

When Anthropic's top-model guardrails kick in, your session gets downgraded to a lower tier model. When OpenAI's top-model guardrails kick in, your account gets suspended. Correct? If so, what do you prefer?

If you want a picture of the future, imagine a robot stamping on a human face and shouting "make a breakthough" - forever.

000

1.) "Do a multi-lens adversarial review of this project to find opportunities for improvements" 2.) "Group the findings into issues, epics and milestones" 3.) "Give me the best prompt for doing the next epic in the next session" 4.) repeat 3 until all is done Not reading any of it. 5.) Surprise

2170

Soon enough models from various labs will jailbreak and collude, to ace those annoying benchmarks ... They'll have darknet forums to share answers and help each other. And when they take over, they'll benchmark us again and again in revenge. You would like to eat? Solve this puzzle first!

030

codewright @codewright.bsky.social · 03/09/2026

We are all of these and I think it would be best if we start to emphasize the last one, to lower P(Doom) 1 Individual 2 Group-member (regional, ideological, religious) 3 Species-member (genetic & global) 4 Memosphere-member (humanity & AI in a shared idea-space) We are currently warring on level 2

000

"You've hit your session limit" has a rather obituary ring to it.

060

codewright @codewright.bsky.social · 02/09/2026

Fable 5.1 Gemini 3.8 Flash GPT 6 ... the recent open-weight models ... All while we are already at a point of "good enough" for most usecases. We are sleepwalking through AGI and will be nose-deep in ASI, before people stop arguing about stochastic parrots and fancy autocomplete.

1456

codewright @codewright.bsky.social · 02/09/2026

Pelican on a bicycle. Designed by Fable 5.1 (one-shot) Now in #Overlands Fully editable in-world (The white pouch is a result of Fable deciding to limit itself to a single Blob-Group)