In their example it doesn't matter is this case if the models get better or not. It matters whether inference gets cheaper to the point that we can afford to ba

In their example it doesn't matter is this case if the models get better or not. It matters whether inference gets cheaper to the point that we can afford to basically throw huge amounts of tokens at exploring the problem space. Further model improvements would be a bonus, but it's not required for us to get much further.