Are we simply our patterns? That question used to belong to philosophy. Now it belongs to the workplace. A fascinating piece from MIT Technology Review and Caiwei Chen documents the viral GitHub project in China called "Colleague Skill". The project was built as a spoof, it imports chat histories and workplace documents from apps like Lark and DingTalk (Slack / Teams equivalents), then generates “manuals” describing a coworker’s tasks, habits, quirks, and work style so an AI

A few months ago, I joined the MIT AI Negotiation Competition. Humans and AI went head-to-head in sophisticated bargaining scenarios. This week, Anthropic published a fascinating experiment, called "Project Deal". Anthropic recruited 69 of their own employees, gave each a $100 budget, and asked Claude to interview them about what they wanted to buy and sell from their personal belongings. They then released the AI agents into a Slack-based marketplace with no human interventi

Every few weeks, a new model takes the top spot. But the ranking is only one lens. The real signals are scattered across a dozen benchmarks → coding agents, computer use, long-context reasoning, hallucination rates, tool-calling accuracy, cost per token, etc. Each one tells a different story about what these models can actually do. What I'd watch is shape of the frontier. The same strengths that make GPT-5.5 great at coding also make it powerful for everyday work on a compute

Michael Crichton noticed something strange about reading newspapers. You open the paper, read an article on a subject you know deeply, and see that it is full of errors. You turn the page, read the next article on a subject you do not know, and trust it completely. Crichton named the effect “Gell-Mann Amnesia” after his friend Murray Gell-Mann, the Nobel physicist who theorized quarks. Somehow you forgot what you just learned one page ago. Are we doing the same thing with AI

Last Sunday in Beijing, an autonomous humanoid robot named Lightning finished a half-marathon in 50 minutes and 26 seconds. The robot ran 3X faster than last year’s model.For context, I run half-marathons, and it would have lapped me twice, easily. It is easy to dismiss this as a stunt. A machine on a dedicated lane, helped up by technicians when it falls, is not competing with human athletes in any real sense. BUT if you are watching this space for signals, you look at the u

"Every Building You’ve Ever Been In Was Designed By Software Built in 1997" - a16z We are standing at the edge of AI’s most massive leap yet: breaking out of the digital sandbox and into the physical world. Between Fei-Fei Li’s World Labs, autonomous vehicles, and the billions flooding into robotics, the race to build physical and spatial intelligence is accelerating fast. There is one massive bottleneck: data from the physical world. Scraping the internet won’t teach a neura

Growing up, I was captivated by stories of grandmasters playing chess without a board. They held a universe of sixty-four squares and thousands of branching possibilities entirely in their minds. It was my first realization that true expertise is the ability to build a working model of reality inside your head. To measure this invisible architecture of the mind, Arpad Elo gave us the "Elo rating." You play, you win or lose, and your Elo rating adjusts. A casual club player si

In the 1950s, Rosalind Franklin produced Photo 51. This was an X-ray diffraction image of a clear "X" shape that provided the crucial evidence for the helical structure of DNA. This week, OpenAI introduced GPT-Rosalind, a reasoning model built specifically for life sciences research and named in her honor. There was a fascinating example in the launch. In a partnership with Dyno Therapeutics, GPT-Rosalind was tested on RNA sequence-to-function prediction using unpublished seq

This week, Peter Diamandis asked a provocative question, roughly: What happens when your customer is smarter than you? Picture a review meeting. Your team spent three weeks on a strategy deck. Your stakeholder pulls up their own AI-generated analysis. It’s more current. It includes variables you hadn’t considered. It reaches different conclusions, backed by logic that looks just as rigorous as yours. The instinct is to panic: the person across the table is smarter than me. Bu