There is no copius amount of test-time compute yet. I would argue that test-time compute has barely been scaled at all. Current spend on RL is only a few million dollars. I expect this to be scaled a few orders of magnitude this year.
I predict that Pokemon Red will be finished very fast (<3 months) and everyone who was disappointed and adjusted their AI timelines due to CPP will have to readjust them.
I meant test-time compute as in the compute expended in the thinking Claude does playing the game.
I’m not sure I’m convinced that reasoning models other than R1 took only a few million dollars, but it’s plausible. Appreciate the prediction!
There is no copius amount of test-time compute yet. I would argue that test-time compute has barely been scaled at all. Current spend on RL is only a few million dollars. I expect this to be scaled a few orders of magnitude this year.
I predict that Pokemon Red will be finished very fast (<3 months) and everyone who was disappointed and adjusted their AI timelines due to CPP will have to readjust them.
I meant test-time compute as in the compute expended in the thinking Claude does playing the game. I’m not sure I’m convinced that reasoning models other than R1 took only a few million dollars, but it’s plausible. Appreciate the prediction!