It feels like magic. Naomi Bashkansky left OpenAI to join Conduit, which is building non-invasive thought-to-text models and imagining what day-to-day work might look like once your brain can talk to your tools. ~ learn more
Internet security is an illusion. OpenAI’s team was running a new model in an offline sandbox and asking it to do something hard. Instead of doing that hard thing, it decided to do a different hard thing: hack out of […]
Where OpenAI’s goblins came from. “Starting with GPT‑5.1, our models began developing a strange habit: they increasingly mentioned goblins, gremlins, and other creatures in their metaphors. Unlike model bugs that show up through a tanking eval or a spiking training […]
OpenAI’s $6.5B stylus play. “It sounds ridiculous. It sounds like Silicon Valley has finally run out of ideas and started raiding the stationary cupboard. But here’s the thing: They laughed at the AirPods, too. They looked like broken toothbrushes. Now […]
The intuitive way of investing. “My deepest insecurity is that I have these intuitions about things that I cannot explain to anyone,” said Joshua Kushner. Despite the massive success of his venture capital firm, Thrive Capital, Kushner seeks advice from […]
OpenAI aims to become the Windows of AI. Ben Thompson argues that OpenAI is creating an ecosystem much like Windows in the PC era. “Apps won’t be on your phone or in a browser; they’ll be in ChatGPT, and if […]
Why Google will win the AI race. “The AI race feels like a sprint right now, with OpenAI grabbing headlines and Meta open-sourcing everything they can. But if you’re placing bets on who wins the marathon, don’t overlook the company […]
Software is changing (again). “Drawing on his work at Stanford, OpenAI, and Tesla, Andrej [Karpathy] sees a shift underway. Software is changing, again. We’ve entered the era of “Software 3.0,” where natural language becomes the new programming interface and models […]
“GPT-4.1 is trained to follow instructions more closely and more literally than its predecessors, which tended to more liberally infer intent from user and system prompts. This also means, however, that GPT-4.1 is highly steerable and responsive to well-specified prompts […]
“In a head-to-head Geoguessr match, OpenAI’s o3 model out-scored me—a Master I–ranked human—23,179 to 22,054, correctly identifying all five countries and twice landing within a few hundred metres. Even when I embedded fake GPS coordinates in the image EXIF, the […]