the argument in the paper seems to be that coding ability is what leads to the tipping point. Eventually human-level (then superhuman) coders augment the AI research process until an AI research agent is developed, and it's exponential from there.
We know they are developing more advanced models, and we know they're secretive about it, but how advanced?... ¯\_(ツ)_/¯
We know they are developing more advanced models, and we know they're secretive about it, but how advanced?... ¯\_(ツ)_/¯