Podcasts

I’ve said before that the 𝗳𝘂𝘁𝘂𝗿𝗲 𝗶𝘀 𝗶𝗻 𝘀𝘂𝗽𝗲𝗿𝗽𝗼𝘀𝗶𝘁𝗶𝗼𝗻. We are watching multiple possible futures exist at the same time, and with every new AI capability, their probabilities shift. This week gave us two examples of that happening in real time. 1\ Mathematician Levent Alpöge working with Anthropic Fable 5 produced a counterexample to the Jacobian conjecture. It is an 87-year-old mathematical problem about when certain polynomial maps must be invertible a

“If I have seen further it is by standing on the shoulders of giants.” Newton wrote that in 1675, echoing a centuries-old idea. It remains the simplest explanation for how progress happens: each generation inherits tools that make the next generation’s breakthroughs possible. A few days ago, I walked through that idea in physical form. At the Computer History Museum in Mountain View, you can move through the history of modern computing in about ninety minutes. Room-sized main

Bloomberg frames AI as making interviews less trustworthy. A more interesting interpretation is that AI has revealed interviews were never measuring what companies thought they were measuring. Hiring is fundamentally a prediction problem. Companies are trying to answer one question: will this person create value in the role? Interviews have always been an imperfect measurement tool. They capture a combination of ability, communication skills, confidence, preparation, backgrou

“Where is the wisdom we have lost in knowledge? Where is the knowledge we have lost in information?” T.S. Eliot wrote those words nearly a century ago. Those questions are more relevant today than ever. Consider that inside your skull sits a three‑pound system that runs the most sophisticated intelligence we know. It contains roughly 86 billion neurons, more than 100 trillion connections, and operates on about 20 watts of power, less energy than a household light bulb. The br

Chinese startup Moonshot AI is poised to launch Kimi K3, a 2–3 trillion parameter, open‑weight frontier model. Reports suggest it could be at, near and even exceeding Anthropic’s Claude Opus 4.8 on key benchmarks (FT). Early signals suggest not just “bigger,” but smarter scaling: architectural tweaks like attention residuals (smarter internal memory) and more efficient long‑context handling (bigger external memory), coupled with China’s already strong advantage in low‑cost AI

The big banks beat earnings. The more interesting story is what those earnings reveal about the future of work. J.P. Morgan, Bank of America, Goldman Sachs, Citi, and others are embedding AI across banking functions. Over the next several years, I expect 10 major shifts across financial services: 𝟭\ 𝗧𝗵𝗲 𝘂𝗻𝗶𝘁 𝗼𝗳 𝘄𝗼𝗿𝗸 𝘀𝗵𝗶𝗳𝘁𝘀 𝗳𝗿𝗼𝗺 𝘁𝗮𝘀𝗸𝘀 𝘁𝗼 𝘀𝘆𝘀𝘁𝗲𝗺𝘀.The next generation of productivity will come from workflow architects who can redesign how wor

Thinking Machines Lab released its first model, Inkling. The startup is run by Mira Murati, who was OpenAI's chief technology officer, and for a few days in November 2023, its chief executive. Inkling features a highly customizable Mixture-of-Experts architecture designed to balance resource efficiency with deep multimodal understanding. It was trained on 45 trillion tokens of text, images, audio. It carries 975 billion parameters but wakes only about 41 billion of them for a

On 16 April 2007, 210 miles above Earth, Sunita Williams taped race bib 14,000 to a treadmill in the Zvezda module of the International Space Station and ran the Boston Marathon. Bungee cords held her in place. The station was moving at about 17,500 miles/hour. She finished in roughly 4 hours and 24 minutes, the first marathon completed in orbit. Sunita Williams was fighting what orbit does to the body. With prolonged exposure to microgravity (weightlessness), weight‑bearing

TL;DR: Last week revealed the true trajectory of AI: intelligent systems designed to compound through feedback. Anthropic restored access to Fable 5, SpaceXAI launched Grok 4.5, OpenAI rolled out GPT-5.6 and ChatGPT Work, and Meta, Moonshot (Kimi K2.7-Code), and Zhipu (GLM) advanced agentic and multimodal AI. The common thread is the intelligence loop: sense, act, measure, learn, improve. The future advantage belongs to whoever compounds intelligence fastest. Last week was th