This is how I’m going to read all references to AI safety going forward. Brilliant.