How does it work? Every meeting of the flemish government in Belgium is live streamed on a youtube channel. When a livestream starts the software is searching for phones and tries to identify a distracted politician. This is done with the help of AI and face recognition. The video of the distracted politician are then…
Marble, our frontier multimodal world model, is available to everyone starting today
Claude Code keeps me waiting. Here I am pressing return like a crack-addicted rodent in a lab. “Yes, I want to make this edit.” I watch as it works, glassy-eyed and bored as the code scrolls by, and on the edge of my seat because my ideas are about to become reality.
I want to present you with two narratives about AI. Both of them are about using this technology to automate computer programming, but they point ... Read more
A new avenue for identifying greedy, badly-behaved bots
Multiple parallel Claude-Code sessions power a self-improving AI factory where agents write, review, and refine code.
New models and new thresholds
Ever wondered how AI agents actually work behind the scenes?
AI can write code, but it can’t build real software. Software engineering remains human work because AI can code, not engineer.
Principal AI Architect. Creator of open-strix, a harness for building agent teams. Writing about AI architecture, stateful agents, and what happens when you give AI memory.
I shared a controversial take the other day at an event and I decided to write it down in a longer format: I’m afraid AI won't give us a compressed 21st century.
Welcome to Internet and Technology Addicts Anonymous! We’re glad you’ve found us, and we hope our community can be as helpful to you as it has been for us. ITAA is a Twelve-Step fellowship of individuals who support each other in recovering from internet and technology addiction. This includes social media addiction…
Gmail doesn't just offer to write your emails for you, they actually do it, and it's up to you to delete the text it wrote. Hard to make a screen shot to demo without revealing personal i…
Large Language Models (LLMs) are typically presumed to process context uniformly—that is, the model should handle the 10,000th token just as reliably as the 100th. However, in practice, this assumption does not hold. We observe that model performance varies significantly as input length changes, even on simple tasks…
Claude Opus 4 and 4.1 can now end a rare subset of conversations. An update on our exploratory research on model welfare.
A field guide to responsible AI-assisted development
One of my favorite AI dev products today is Full Line Code Completion in PyCharm (bundled with the IDE since late 2023). It’s extremely well-thought out,...
I have previously blogged about the relatively new trend of AI slop in vulnerability reports submitted to curl and how it hurts and exhausts us. This trend does not seem to slow down. On the contrary, it seems that we have recently not only received more AI slop but also more human slop. The latter … Continue reading…
This Gradient Updates issue goes over the major changes that went into DeepSeek's most recent model.
OpenAI and Nvidia are both under threat from Google; I like OpenAI’s chances best, but they need an advertising model to beat Google as an Aggregator.
Pure scaling in shambles
Some (well-known) consequences of AI automating work
And a plea to think seriously about climate change without getting distracted
How the Elixir community can survive — and thrive — in an age of LLMs.
This post is about how we've tried to develop a position on AI in the context of learning and growth at RC.
Finally, we train an LLM! The final part of Chapter 5 of Build an LLM (from Scratch) runs the model on real text, then loads OpenAI’s GPT-2 weights for comparison.
Everyone's using AI wrong. Its superpower isn't writing for you. It's reading everything you've ever written and finding patterns you missed.
Today, we’re introducing WorldGen: a state-of-the-art end-to-end system for generating interactive and navigable 3D worlds from a single text prompt.
... in specific domains. Three case studies with GPT-4o, o1, and Claude Sonnet 3.5, and what they mean
