All of my long-form thoughts on AI, robotics, and their interface with society, collected in chronological order.
Anthropic's protein-design system runs on a 16,000-word prompt, and only a third of it is about proteins. OpenAI's proposed Navier-Stokes proof now shows a similar process, alongside harder questions about validation and provenance.
A team can get faster, flatter, and happier while quietly losing the ability to decide how it works. Notes on how to tell whether your team still authors its own arrangement, through Guattari's idea of transversality, and what to do about it.
Foundation models are eating software from both ends. Seven observations on where the value actually lives in vertical AI, and whether AI-native services companies really are the future: why owning the outcome beats owning the workflow, why mandatory gates are a real moat and which five traditional software moats are already gone.
Notes from the field on what actually works when building AI-native products. The SaaS playbook is breaking down: the product is now a substrate, not a solution. Observations on omnipresent interfaces, flat knowledge, closed loops, and displaced bottlenecks.
In the span of ten days, AI agents built themselves a social network, a marketplace, and a gig economy platform. Is this a new universe, or just an elaborate prank?
The core challenge of deploying automation in a legacy environment, like a warehouse, isn't the robot, it's the interfaces. When systems with different natural speeds are forced through shared integration points, small disturbances amplify instead of dampening.
Twelve falsifiable predictions for embodied AI in 2026, from VLA scaling laws to humanoid safety incidents. The models are getting good enough; the deployment wall is the real test.
Why do different AI models keep giving the same creative answers? A NeurIPS Best Paper study shows open-ended responses collapse into a few dominant modes across models and even across repeated samples. Let's explore what that homogenization means for culture, and how to push for more diversity in practice.
Silicon Valley doesn't just shape the world, it rules it. The tech giants that call it home have amassed extraordinary influence over political and social structures, with decisions that ripple far beyond U.S. borders. Europe is caught in a bind: should it regulate, imitate, or innovate?
The benchmarks driving AI development have a fatal flaw: they reward performance over competence, eloquence over accuracy. Understanding this gap is the first step to building, and choosing, systems that actually work.
AI systems are suddenly capable of tasks they consistently failed at months ago. The secret isn't just about bigger models or more computing power. Behind the latest breakthroughs is a quiet revolution: companies have stopped training AI and started teaching it.
Post-training is all the rage. Are we going from training to teaching? What does a curriculum look like for LLMs? Let's find out.
MCP is fast becoming the "USB-C for AI," connecting models to the real world. This post breaks down what it is, why it matters, and where it's headed.
Why Your AI Agent Might Be Flying Blind - Would you trust a robot with no sensors? Then why are we building AI agents that way?
After nearly three years building Claryo from conversations on commutes into a truly awesome AI product, I've decided the time is right for me to step away and begin a new chapter.
When OpenAI refused to publish GPT-2 for fear of its malicious use: an analysis of the challenges of automatic text processing, between linguistic regularities and scientific responsibility.
How do machine learning algorithms redefine intellectual property? Between patents, trade secrets, and attribution questions, an exploration of the legal and ethical issues of AI.
The Portrait of Edmond Bellamy sold at Christie's for 480,000 CHF: an exploration of GANs, generative art, and fundamental questions about who is truly the artist when a machine creates.
From Boston Dynamics' dancing robot to fundamental questions about artificial intelligence: an exploration of the "uncanny valley" and the current limits of machine learning.
Programming and comanding grasping robots by demonstration is notoriously difficult due to the number of degrees of freedom involved. In this work, we propose an intuitive, Augemented Reality method where the user is able to 'point and click' to a goal pose for the robot end-effector to go to.
Programming and comanding grasping robots by demonstration is notoriously difficult due to the number of degrees of freedom involved. In this work, we propose an intuitive, Augemented Reality method where the user is able to 'point and click' to a goal pose for the robot end-effector to go to.
The Swiss confederation provides a large amount of data spanning virtually every aspect of the political process in Switzerland. With a convenient API to query the proceedings of committees, councillors, parties, logs of voting sessions … From this massive resource we decided to study voting patterns in the Swiss government.
What makes Starry Night look like a van Gogh painting ? Is it the colours ? Is the brush stroke ? Is it the subject ? Somehow when we see this painting we associate its style to its author. If the style of the painting can be thought of as an independent component from the subject it represents, would it be possible to apply the style of *Starry Night* to another painting of picture ?
An example of approximative scientific vulgarization. The article is really bad and misleading, the brain does not encode 11 dimensions. What the paper is about is the ability to find the depth and distribution of millions of subnetworks of neurons within the brain.