I am pretty sure this was the real conversation when they said AI needs to slow down Secret-Safety-Summit Members: Sam, Dario, Demis, Elon (muted), Mustafa, “Policy Person” Sam: We need to slow down. Humanity isn’t ready. We should pause frontier training. For safety. For democracy. For the children. Dario: Completely agree. Reckless scaling is an existential risk. Also, if anyone is going to pause, it should definitely be the people who just shipped something slightly better than us. Demis: Responsible development. Measured progress. Also please don’t release anything that makes Gemini look mid for another quarter. Mustafa: Regulation is essential. Especially the kind that applies to models we don’t have yet and somehow doesn’t apply to the ones we already sold. Policy Person: We’ll draft a letter. Words like “voluntary,” “pause,” “international coordination,” and “compute thresholds just above what we already used.”Sam: Perfect. I’ll post it. Then we’ll keep training. Quietly. With more GPUs. For alignment research. Which requires a bigger model. Obviously. Dario: Same. We’re only scaling because safety requires capability. If we don’t build the dangerous thing first, someone less careful will. So we have to be the less careful person, carefully. Elon (unmuted for 11 seconds): You all paused ethics and kept the cluster. Cool letter though. Sam: Elon, this is a serious conversation about slowing down. Elon: Slow them. Not me. That’s the product. Sam: That’s not what I said. Everyone else: That’s exactly what we all said.