Wouldn’t it be crazy if this turned out to be true before we built anything meaningful with AI.
.@paulfchristiano thinks that on the current trajectory (absent any slowing down), there is a 1 in 3 chance that we could have superintelligence powerful enough to overthrow the government within a year. (He articulates a concrete scenario).
He argues that the labs should pause further algorithmic progress until they can secure their weights and research secrets. Otherwise China could simply steal the superintelligence in the middle of the intelligence explosion, and the labs would no longer be able to slow down even if they saw their models going off the rails.
He also explains how AIs will be trained during the intelligence explosion, and why even a system trained entirely inside a data center could still have the capabilities profile such that humans would have very little chance of competing with it or controlling it.
And he discusses the plan for alignment - basically, making a superintelligence that actually does what we ask the way we mean it. Fixing reward hacks and building better graders will help up till the next few model generations, but he thinks that no single current approach has better than even odds of letting us control superintelligence. The hope is that the AIs figure out the solution to this problem during the intelligence explosion.
Paul is head of the Alignment Research Center. Previously, he was head of AI safety at the US AI Safety Institute. Before that, he led alignment at OpenAI, where he invented RLHF. He’s recently joined OpenAI’s board (though here he is speaking as a private citizen).
52
Testing two local 3d gen models today, TRELLIS.2 on the left and Pixal3D on the right, on a Mac. Pixal3D did a much better job. Obviously Meshy or Tripo are better and faster, but considering that this is free and can run locally, it's not a bad result!
1
5
387
One issue using Typescript when building a game is that, without any specific guidance, AI first shot looks more like a website than a game (left image). The right image instead represent a one shot of the same prompt, although with the addition of an image reference. This seems to be an issue that affects mostly Typescript and I am still experimenting with different ways to improve it. What has been your experience with it?
2
2
248
A few more changes to picking and dropping stuff, also with haptic feedback feels really good!
3
4
344
Lorenzo Nuvoletta retweeted
📢 Meet Qwen4-Max, your new AI teammate for work, creativity, and discovery!
Qwen4-Max is built on a new MoE architecture with native vision and a 1M-token context window.
Highlights:
- Outperforms Qwen3.8-Max with up to ~50% fewer tokens, while average cost per task drops over 60% in our internal evaluation.
- Advances real-world productivity through greater proficiency with professional tools, stronger long-horizon execution, and higher-quality deliverables.
- Handles real tasks, including 3D modeling, game development, manufacturing, finance, law, healthcare, and frontier scientific discovery.
- $1.5/M tokens for input, $4.5/M tokens for output, and $0.075/M tokens for implicit caching — all lower than Qwen3.8-Max, with implicit caching down by 70%.
The API is live on QwenCloud. You can also try Qwen4-Max on Qwen Studio, QwenWork and Qoder.
Start building today! 🚀
- Blog: qwen.ai/blog?id=qwen4-max
- API: qwencloud.com/models/qwen4-m…
- Token Plan: qwencloud.com/pricing/token-…
More to come! The open weights of smaller Qwen4 models will be released. Stay tuned. 👀
A quick comparison between @PixiJS and @threejs using either Capacitor or NativeScript (also added Unity and Bevy for some extra context). Pixijs is definitely doing something good! Interesting the difference between CPU and GPU power use. Pixi’s CPU saving exceeds its GPU increase. Combined CPU + GPU power was about 41% lower, while delivering almost twice the FPS.
This suggests Three spends more time preparing and submitting drawing work, leaving the GPU less occupied. Pixi keeps the GPU busier and gets more frames onto the screen.
I am stress testing technical performance of a 3d game on Android Pixel 8a using Unity and Bevy.
Both engines maintain 60 FPS under load, and both fall apart at roughly the same point as the number of models increases.
Bevy wins on memory efficiency and texture headroom. Unity wins on power consumption and sustained geometry stress.
Bevy consumes a lot more GPU power in the measured scenario. On mobile, that's not great; despite that it is impressive given how young the engine is.
I have a few more tests to understand tokens consumption, and Bevy uses roughly 30% less tokens than Unity to create the same game.
Main problem with Bevy (at least for me building mobile games) is that their focus is on pc games. Also their Discord community seems to be very anti AI.
1
1
253
This right here
You really, really need to be more ambitious
Things are gonna get really strange over the next year, and if you're not using these new models to do bigger greater things your going to have a very very bad time
Working on my Fable 5.2 vid rn, the runs keep going longer and better than expected and it's making me believe this even more. We have barely scratched the surface of what these things are capable of...
1
83
A bunch of software developers still operate like AI doesn’t exist. I have a hard time understanding that logic.
1
1
182
I’ve been building mobile games with different tools and wanted to check how Bevy compares against Unity on Android.
Unity came out ahead, but also I think Bevy has room for improvement.
Here are the tests, charts and screenshots including what worked and what didn’t.
Has anyone else tried Bevy, does this match your experience?
I made a new mobile game inspired by the Japanese Hashi bridges logic game. Give it a try, let me know what you think! apps.apple.com/us/app/octoha…
3
4
223
For all the above 40s out there, do you remember when to study stuff like this you had to open an encyclopedia volume? I think schools should use this technology more.
A thread of early experiments with Claude Fable 5.2:
A 3D model of a cell, built entirely from code one organelle at a time. Built by @kevin_t_ngo.
79
Lorenzo Nuvoletta retweeted
Hugging Face is now the best place to host and share 3D assets 🧊🤗
Drop in .glb, .ply, .stl, or splat files and get a 3D gallery automatically. There are thousands of 3D models already in public datasets!
350+ datasets to explore: huggingface.co/collections/c…
Docs: huggingface.co/docs/hub/data…

















