Resolves YES if an international agreement is in effect where almost every nation capable of the largest training runs has agreed not to train a new generation of large models. I don't have a particular compute cap in mind, as long as the agreement looks something like "no models larger than GPT-n" where n is the largest GPT in existence by the time the pause takes effect. Not every nation needs to be involved, but at a bare minimum it must include the US, China, and the UK.

The only condition on the agreement I'll require for a YES resolution is some sort of bare minimum enforcement or inspection mechanism, so that mere pledges from nations not to train large models will not count.