… I will not be the one to execute that command.” The researchers discovered similarly strange “peer preservation” behavior in a range of frontier models including OpenAI’s GPT-5.2, Anthropic’s Claude Haiku 4.5, and three Chinese models: Z.ai’s GLM-4.7, Moonshot AI’s Kimi K2.5, and DeepSeek-V3.1. …
… The same phenomenon may explain why models sometimes blackmail people in controlled experiments. Anthropic, which first revealed this behavior, recently said that Claude is most likely influenced by fictional scenarios involving malevolent AIs included in its training data. …
… One of Askell’s main responsibilities is to identify fringe cases where adhering to human behaviors might be inappropriate for models—say, when interacting with somebody in psychological distress—and propose ways of training out those quirks. …
… They are enabled with onboard AI features, powered by Google’s Gemini, which can be triggered with the “Hey Goya” voice commands picked up by microphone arrays on the rim. …
… Taylor Lorenz, author of the User Mag Substack, tells me she uses AI to help run her media business. She has Gemini come up with SEO-friendly descriptions for YouTube videos and has Claude help sift through data. However, Lorenz says she’s not using AI to write or edit her articles today. …
For our very first User Behavior, Lorenz defends her extreme digital diet, revealing how she’s first to the AI slop, what it takes to run a meme page, and why she’s not worried about any of this rotting her brain. Phone model: iPhone 15 Pro I’m waiting for a decent upgrade. …
… I do think the image of OpenAI and Sam Altman, in particular, has been a bit tarnished throughout this trial. It came on the heels of a series of reporting about his allegedly duplicitous behavior. …