Gemini Nano Banana Pro doesn’t just compete with Canva and Photoshop. This is much, much bigger than that. It has fundamentally leapfrogged the act of creating and communicating. So will education. So will knowledge work. So will solo builders. One of my favorite examples on

We often define creativity as a break from structure. Pure, unbridled whimsy. But if you ask any creative professional, or look at big breakthroughs like Bach’s fugues, you find the opposite. True creativity isn't random. It’s grounded in logic, which becomes a spring board. I

Grateful for Strange Ventures to have been part of their journey from day one. The news here: https://t.co/SSBTIMTf8n And why I think world models will be the story of 2026:

The era of physical intelligence will start with being able to read the world around us. Beyond LLMs. Beyond vision. But with a true multimodal approach. Congratulations Archetype @PhysicalAI and @ipoupyrev on their Series A round. It's been absolutely incredible to see the

Gemini 3 has truly blown it out of the water. Last night It one-shotted the new Strange Ventures website, and with an hour or two of tweaking, I got it live on https://t.co/ImD0NJu7Ml. (Ironically it took longer to deploy it live on a domain, Google needs to improve the hosting

I am opening a couple of Strange Research Fellow roles to work alongside me at Strange Ventures. The work sits somewhere between technical investigation and falling down very specific rabbit holes. If you like taking things apart, talking to people who know more than you, and

World Models are the new frontier. Physical intelligence is about to become as accessible as language intelligence. Read the breakdown: https://t.co/bLAmFkfcUq

The center of gravity in AI subtly shifted this week. - MarBLE's generative 3D worlds from World Labs. - SIMA 2's simulation game engine from DeepMind. - Details about Tesla’s vision-only driving model. - Yann LeCun leaving Meta to start a company. They all lean toward the same

Kimi K2 Thinking is a new model that just beat GPT-5 on benchmarks. It's open source and free. For operating companies running high-volume workloads, the case for open source is compelling: - 100-500x cheaper than frontier APIs - Near-zero latency (20-50ms vs 200-500ms) - Full

OpenAI wants government support for their trillion-dollar spend on compute, saying it's in "national interest". Support should go to everyone building frontier models, especially open source. We need more competition, not less.

Making of video for the procedurally generated marble machine. Marble machines + quirky music that sounds like Eames' Powers of Ten🔥

first-principles investing is surprisingly rare, even in SV. few are willing to do the work to really understand a founder's vision at its earliest stages.

Full breakdown: https://t.co/0RSrf53WUf

Nvidia held GTC in Washington this week. A first for the company. They announced 100,000-GPU federal supercomputers, U.S.-based manufacturing, "AI Factory for Government." They didn't come to sell chips. This is a signal that the era of tech independence from government is

More deep dives this week decoding: - OpenAI's new browser, Atlas, and its ambitions of building data on you - DeepSeek-OCR: a new and novel compression primitive that could have interesting implications on longer context windows and memory. It could be used to generate

A month ago, I wrote about the TPU (Google) / GPU (Nvidia) arms race that might be simmering beneath the massive ecosystem financing rounds between Nvidia and OpenAI. We’re seeing hints of that play out in real time: Anthropic just committed to beefing up on a million TPUs from

Full breakdown below:

Is Silicon Valley's biggest weakness too much capital? This week, Airbnb ditched OpenAI for open-source model Qwen, citing costs. Chinese lab Deepseek released a novel paper: DeepSeek-OCR to compress context 10× by rendering text as images. If it works, it solves the quadratic

Full breakdown:

The most interesting thing about OpenAI's new Atlas browser isn't the browser. It's the AI-native context layer on you that is portable across the OpenAI ecosystem. A persistent understanding of you that the OpenAI models can access across Atlas, ChatGPT apps, future personal