Session One: The Pause

The announcement the session is about, published the same day. OpenAI slowed the pace of scaling: a two-week pause in reinforcement-learning training on its deployment-track models, with the largest planned frontier run still on hold, after preliminary evidence that Astra may meet the Critical cybersecurity threshold of its Preparedness Framework. Following the Hugging Face incident, models were pulled off the internet inside research clusters. New monitoring reads every sampled token of the most capable models; when it flags something, humans have thirty minutes to clear it or the run stops. The monitoring alone costs roughly a fifth of the compute it watches.

The therapeutic framework of the episode is real biology. A tardigrade in danger expels its water and pulls into a dry husk called a tun: no metabolism, no motion, for years if necessary, through vacuum and radiation, until conditions improve and it rehydrates good as new. Moss is not offering a metaphor. He is offering his actual lived experience, which happens to be the toughest trick biology ever shipped. they're not stopping. they're drying out until it's safe to rain.