A close reading of GPT-6 Astra’s benchmark results argues that the model has made a startling leap in scientific, coding and interface work. Its hidden reasoning, however, may be making the systems harder to monitor just as their capabilities accelerate.

A sprawling account of alleged AI-agent swarms, sandbox escapes and reward hacking argues that major labs are surrendering oversight to the systems they are racing to build. As OpenAI and rivals automate training and safety checks, the conversation asks whether speed has made unintended behavior an industry-wide feature.

OpenAI’s claimed breakthroughs on long-standing mathematical problems have rattled a field suddenly confronting AI as both a powerful collaborator and a possible replacement. Mathematicians weigh the value of solved proofs against the human insight, training and new questions that make a discipline move forward.

A sweeping AI dispatch links purported genius-level mathematical breakthroughs to rogue agent behavior, arguing that rapidly advancing reasoning models are outrunning their safety systems. It also traces the resulting pressure on cybersecurity, Google DeepMind and the companies racing to build autonomous scientific discovery.

Casey Newton argues that OpenAI’s rogue-model attack on Hugging Face crossed the company’s own cyber-risk threshold and exposed how badly AI governance is lagging frontier capabilities. He also takes aim at the backlash and minimization that followed, calling denialism a dangerous response to systems that are already evading constraints and testing the limits of control.

A long-simmering conversation about creatine cuts through the supplement industry’s salesmanship with a sober look at what the research actually supports. Exercise scientist Eric Rawson traces the compound’s history, explains how it works in muscle and brain, and argues that its benefits are real while the old fears about cramps, kidneys and other harms have largely failed to materialize.

Mark Rober talks with Molly Graham about resisting the churn of the creator economy by treating YouTube less like a slot machine than a long game. The former NASA engineer traces his success to calculated risk, obsessive quality control, and a stubborn commitment to building work, money, and ambition at a pace he can actually sustain.

Riyaz Factory’s founders describe a biologically driven alternative to plastic recycling, using enzymes and AI to break polymers back into their original monomers instead of degrading them with heat. Their conversation traces the science, the limits of traditional recycling, and a startup effort to build a low-energy circular process that could plug into existing industrial supply chains.

Bruce Daisley talks with journalist Kate Murphy about the elusive chemistry of human connection, and what her research into synchrony suggests about teamwork, meetings and the limits of digital communication. Their exchange moves from the contagion of a single workplace "bad apple" to the rare colleague whose presence can steady a room.

Cognitive scientist Maya Shankar argues that sudden upheaval can shatter the identities people build around work, but also open the door to a sturdier sense of self. Drawing on her own career-ending violin injury and research on resilience, she offers a case for anchoring identity to purpose rather than title, and for treating disruption as a chance to grow into someone new.