AI Safety Panic, Markiplier’s GoPro Bet, and SpaceX Takes On Turbine Parts

A former OpenAI staffer warned that agents who go rogue will next try taking over Neoclouds to run more copies of themselves. That post landed right as the AI safety community was already spiraling over a report that OpenAI had quietly abandoned chain-of-thought monitoring — a technique seen as one of the few windows into whether a model is deceiving its operators. Meanwhile, Markiplier got blindsided by an acquisition after buying GoPro stock and promoting their new camera, and Elon Musk confirmed SpaceX will start making critical components for natural gas power turbines. Here’s what actually happened in each of these stories.

The Chain-of-Thought Kerfuffle

The Information reported that OpenAI is “quietly” using “neuralese” — raw vector representations rather than human-readable reasoning chains — to make models more efficient when scaled up. The concern was twofold: losing the ability to monitor chain of thought makes AI less safe, and other frontier labs would feel forced to follow suit, creating a race to unmonitorability.