I spent 20,000 dollars in API credits having different AI models play each other at Blood on the Clocktower (a Mafia or Werewolf type social deception game). The results were fascinating. Models have gotten much better over the course of this year at creative on-the-spot scheming and bluffing, and also at reasoning as a team to break apart these underground conspiracies. I go through some of the most standout games these AIs played in this video. The new models are getting pretty good. If my friends pulled off some of these moves while we were playing, I’d give them a fist bump. Fwiw, OpenAI’s newer models are much better than Anthropic’s at this game. You can see the full benchmark at the website, and I’ve also made it so that you and your friends can play again the AIs:

Is my perception off or is Apple just lying here? Anytime I build for or use Safari, it’s slower, has fewer features, weird bugs and plenty websites just don’t work. It’s a fight to get even somewhat close to Chrome’s performance.

we're giving you some RAM back. local threads could grow memory usage as they became long, even without starting other processes. @btraut declared war on the problem. enjoy in equally important news, @matteing made his handwriting into a font, which i use below

Do individual limits still make sense? As capabilities increase you want to work with and support an agent that represents your project, that runs loops and evolves your codebase. The humans merely curate the automations.

When our team first moved to multiplayer sessions, where everyone can read each others prompts, I was initially very hesitant I say a LOT of dumb things to my agent That is still true, but I've learned two things since gaining full visibility into all the sessions of my teammates: 1. Everybody else says dumb things too 2. There is a ton to learn from how other people on the team drive their agents For example, I wanted to learn how our UI/UX guy, @vhbrzezowski , prototypes with agents - so I just asked! My agent read all of his sessions and helped me understand Vyctor's workflow using our `show_widget` tool to iterate on mockups built from our actual frontend code Once you get over the fear of sharing your prompts with others it tears down so many useless knowledge silos across your whole company.

Kimi K3.1 recalls instructions it was never given on Terminal Bench. They match the hidden tests. K3.1 while solving: "Instructions likely distilled pitfalls from previous agents." 🤯 When we edit a task to contradict them, K3.1 refuses the new instructions and fails. Other models follow the edit. Is it terminal(ly) benchmaxxed? 🧵

another brutal reality check for the open-sauce china shills GPT-5.5 released on April 23 outscores all tested chinese models and OpenAI and Anthropic have been holding back their strongest models since the supply chain risk designation, so we aren't even seeing the full extent of the domination