Develop AI systems to efficiently learn new skills and solve open-ended problems, rather than depend exclusively on AI systems trained with extensive datasets
Discover how this asset is connected to other assets on the platform
Throwing down some ideas and I had this morning.
One of the ideas that stuck out was some concept of an evolving search space.
There's some kind of initial filtering we do to narrow down what kind of problem we're working on. A quick scan of the input and outputs let's us know if it's a recoloring, translation, pattern extraction, or something else which we might not have seen before.
We might use this initial classification to narrow down the search space. For example, you don't need to test any translation subroutines when the puzzle is purely a recoloring exercise.
Additionally, there might be some idea that as we search for a DAG of operations, the search space may be changing, either directly via adding new subroutines to try, or creating new ones (LLM generated?).
As you get further into solving the puzzle, you might need a different set of subroutines to finish the job. If we can somehow be smart about which sets we search through, the faster we can search.
Theoretically, it should still be possible to search with all subroutines as candidates, but that's likely to be too large to accomplish in a reasonable amount of time. The challenge does have a time limit.
Discover assets like this one.