Model training has grown complex enough to require many specialised stages and teams, and manual coordination between them doesn’t scale. At Aleph Alpha, we’ve built Savanna, a model factory that implements the entire training pipeline in code.
Cunyan Ma’s interactive tool exposes the power structure that extracts low-wage labor from data workers in the Global South to make everyday AI algorithms function.
Wilson Lin at Cursor has been doing some experiments to see how far you can push a large fleet of "autonomous" coding agents: This post describes what we've learned from …
A configuration mismatch in GLM 5.3 on Workers AI triggered an SGLang bug that exposed KV cache pages across requests. Here is how we addressed the issue.
How I burned a full Claude limit in 30 minutes and built my own deep research pipeline instead: 3 subscriptions, shared memory, a clear role for each model. You can build the same from what you already pay for.
On quitting, the spread of AI, and the loss of an ideal.
You are relying very heavily on the priors of the model to do a good job for you. What happens when you cannot judge whether those priors are good?
Coding in Rust and others; making coding videos.
My keynote at ICML 2026
We’re extending Project Glasswing to approximately 150 new organizations in more than fifteen countries
When OpenAI released their proof that solutions to the Navier-Stokes equations can blow up in finite time, they also released a formal proof in Lean 4.
Keep track of artifacts, not scratch. Alt title: Claude, please stop trying to memorize random crap
Against the Quantification of Integrity When the measure of language becomes its target, it ceases to be good language. 💡Nerd Rating: 1/5. I discuss the origins of certain linguistic tics in LLMs and what it means for writing, student assessment, and thinking. "It's not x, it's y.
Today, we're releasing Dream-1, a World Action Model that controls robots it has never seen, performs tasks it was never shown, and runs from $3/hour.
“Preface to the highest stakes negotiations in history.”
NVIDIA today launched the NVIDIA Vera CPU, the world’s first processor purpose-built for the age of agentic AI and reinforcement learning — delivering results with twice the efficiency and 50% faster than traditional rack-scale CPUs.
I've been using AI fairly heavily since last November and the whole thing is a funny experience . An agent will do something that, if a human did it, you'd immediately fire them. My reaction, of course, is to act as if this is great and spin up a thousand agents so they can do even more of that.
Sarvam open-sources 30B and 105B reasoning models trained in India: MoE LLMs with leading Indian-language benchmarks; weights on AI Kosh and Hugging Face.
Political bias in AI measures where every major AI model stands on charged political and ethical questions: run many times, no web search, plotted with error...
We used agents to port all of Git to library-first, memory-safe Rust and it passes the entire C Git test suite.
Edit: Woah, this took off! If you want more details you can subscribe at the bottom of the post or on YouTube. I’m working on a more detailed highlights/commentary post and video. Previously: We got Claude Code playing Factorio, Towards beating Factorio autonomously, and Faketorio: an agent testbed I got GPT-5.6 Sol…
A safety-filtered pretrain can duck a charged word without refusing. It puts a fraction of the probability an open-data pretrain puts there. We call that...
At the recent EMFCamp, I bumped in to Professor Kate Devlin who was talking a little about the AI Resist List. I think one of the useful things we can do is loudly complain about shitty uses of AI. Here's an example. As mentioned in a previous post, I'm contacting companies which send malformed calendar invites. One…
Astra is the first OpenAI model to meet the Critical cybersecurity capability threshold under the Preparedness Framework, with stronger safeguards for release.
Repository drawgent by yanndegat.tngl.sh on Tangled.
[UPDATE 2026-06-09]: a few months on, I have more misgivings than certainties about this.
38 zero-day security vulnerabilities, three critical, and the shift from disclosure to prevention in healthcare software
Exploring practical ways to make your game more legible to AI agents and how to make agentic iteration faster.
Solid core + capability-based sandboxes + LLMs = Users with superpowers
