Alex Goodman retweeted
Just released v2 of my Three.js training dataset!
3,445 scenes across 50 domains, including 1,431 voxel scenes.
Hugginface: H
hugginface: huggingface.co/datasets/ccor…
Modelscope:modelscope.ai/datasets/ccore…
Alex Goodman retweeted
I really feel like AI (aside from software dev) is just the ultimate add on for the people who have already been the go to people for knowledge in their circles, but now we’re turned up
the smart are just getting smarter (and richer) and some are missing the wave completely
Alex Goodman retweeted
America needs to have vibrant open weight models to ensure liberty. This will help AI proliferate and decentralize. AI in the hands of everyone is the best way to prevent a dystopic AIpocolypse.
Similar to how the personal computer empowered an individual to go up against a corporation. And how open source software has been key to innovation, competition, and progress. Open weight models are an opposing force against AI centralization and totalitarianism, where everything is controlled and decided by a few large tech monopolies and/or governments.
America and its values depend on AI being an open technology.
Alex Goodman retweeted
In the richest country on Earth, starter homes now cost $1 million in a record 242 U.S. cities.
The median first-time homebuyer is now 40 years old — up from 28 in the early 1990s.
Unacceptable.
Instead of investing in endless wars, we should invest in affordable housing.
Alex Goodman retweeted
Zen is getting DRM support! Very soon you'll be able to watch and listen to your favourite streaming services like Netflix or Spotify directly in Zen!
We know this has been your most anticipated feature, and we're thrilled to finally deliver on it!
DRM support
Alex Goodman retweeted
Unreal Engine 5.9's AnimGen plugin is INSANE! 🤯
You can now generate character animations directly inside Unreal Engine 5.9 using AI, completely FREE on your local GPU!
No subscriptions. No cloud processing. Just text-to-animation directly inside Unreal Engine!
Alex Goodman retweeted
[Day 10/30] Now you can access a full fledged file manager and terminal in Luma's TV mode!
Alex Goodman retweeted
In Neural Networks are Already Clean, @nickcammarata argued that neurons were far more legible than previously thought – and that this was great news for interpretability and safety.
Now, we’re releasing the first technical companion to Nick’s essay. It shares our findings from studying thousands of neurons using our novel automated interpretability pipeline, and outlines why we now think of neurons by their computational function, instead of whether or not they are polysemantic. Here’s what we found: 🧵
interpretability might be the most important scientific problem in the world right now, as I don't think we can solve alignment if we don't know what we're building. I think I have good news about it. I co-authored early work in mechanistic interpretability at OpenAI and came away pessimistic, because neural networks appeared to fold into arbitrary tangled knots. years later, I started openai's first large-scale automated interpretability project, finding about 2% of neurons legible with the primitive tools we had
with modern agentic tools and a much closer look, the picture looks totally different. the vast majority in models up to trillions of parameters are legible, and their circuits beautiful. many famous examples of tangled neurons turned out to be misunderstood. The roadblocks I expected mostly aren't there. I now think ambitious interpretability will be solved by the time we need it most, and possibly quite soon, as automated scientists keep improving. it may be a bitter lesson for interpretability: the simple thing, at absurd scale, works
already-clean.goodfire.com/
Alex Goodman retweeted
Two years ago, I started a side project solo on my RTX 3090 Ti. This year, it became an independent SIGGRAPH paper.
Here’s how I got there: teaching myself ML, doing research, writing, navigating peer review, and advice for anyone taking a similar path.
alexandergoslin.me/writing-a…
Introducing InfiniteDiffusion, my independent paper accepted to #SIGGRAPH2026!
I have one RTX 3090 Ti. No funding, advisors, or team. By day I'm a new grad SWE at Walmart.
The paper has two main contributions:
- InfiniteDiffusion: a new approach to infinite generation with diffusion models.
- Terrain Diffusion: the world’s first learned procedural terrain generator.
Here’s why this matters, and how they are connected. 🧵
Alex Goodman retweeted
Weight space programming on image model : SDXS-512
Agent edited 0.00017% of the network to turn flat postcards into scenes you could walk into. No training.
Recipe to reproduce in comments below.
Alex Goodman retweeted
66% want Medicare for All
61% want a billionaire wealth tax
57% want a $20/hour minimum wage
63% want a data center moratorium
58% want to end military aid to Israel
73% want to ban super PACs
79% want a 10% credit card interest rate cap
No more excuses. Democrats must deliver.
Alex Goodman retweeted
weight space modding is going to unlock so many robotics usecases holy shit
Task: edit Depth Anything V2 Large's weights directly to improve edge detection.
Result : It now finds 41% more of the real depth edges. About 0.35% weights edited.
Weight space harness in action in comments below and how to reproduce (no code needed because mods are THAT easy)
Alex Goodman retweeted
I performed brain surgery on a tiny 135M model. It's score on multiplication problems it had never seen went from 21% to 83% without any training. I grafted artificial neurons directly into the model that changed the way it solves problems.
Alex Goodman retweeted
JUST IN: U.S. homelessness records its largest annual decrease in nearly 20 years.
[1/3] turns out you can teach a 35b model to weight-space program better than opus o_0
earlier this week, @rikiyatakehi and @jayden_teoh_ showed that astra can manually write neural networks to solve atari games, one weight at a time. they call this weight-space programming.
on the @trajectorylabs platform, @o__jerry__o and I post-trained 27b and 35b-a3b qwen models to write networks for atari games. we curated WeightProgramBench, a set of 322 simple games to teach models to manually write neural networks.
we combined 51 atari-style games with 266 games adapted from crisp-game-lib for training, holding out the 5 original minatar games (freeway, breakout, asterix, seaquest, space invaders).
on freeway, qwen-27b climbed from averaging 21.0 to 47.1 points **without ever training on freeway**, but still not quite at astra’s 63.9 points.
Max Headbox
A voice agent that lives on a Raspberry Pi 5 and never calls a cloud model
Max Headbox listens for a wake word, transcribes what you say, and lets a small local model pick a tool to run on the Pi. Nothing leaves the device.
It runs gemma3:1b and gemma4:e2b through Ollama and needs about 6GB free for the models. Tested boards: 8GB and 16GB. Vosk handles the wake word and faster-whisper does the transcription.
A tool is one JavaScript file in src/tools with a name, parameters, a description and the code to run. Mark a tool dangerous: true and the agent asks you YES or NO before it acts.
The backend is now Express.js, so Node 22, Python 3 and Ollama are all you need.
github.com/syxanash/maxheadb…
AI-generated artwork is the most common use dexerto.com/gaming/almost-ha…
You don't need Astra to edit a model's weights! Weight-Space Programming works with open-weight LLMs, too.
I gave GPT-OSS-120B a pretrained LunarLander policy and let it directly modify the network's weights.
→ Reversed thrusters: 0% → 88% landing success
→ Crosswind: +5.9pp on held-out episodes
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math.
It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training.
We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵
Blog: rikiyat.github.io/blog/weigh…
Readers added context they thought people might want to know
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math.
rikiyat.github.io/blog/weight-sp…
github.com/RikiyaT/lm_und…
huggingface.co/Qwen/Qwen3.5-0…
huggingface.co/RikiyaT/Qwen3.…



























