Fable-class models have a larger step-size of problems they can solve.

Their adjacent possible is larger.^[b]

They can handle more of a leap without needing a human's judgment in the loop.

Step-size is about how much effort can you do before you need some form of ground-truthing?

You can ground-truth in the real world, or you can use human judgment as a proxy.

One of the reasons the models are so great at coding is that they can ground-truth their own software in the training loop by executing the code and seeing if the output is what was expected.