> I guess I'm contesting that idea you are putting forward that the data from the games these humans are playing, which are at a vastly lower level that the top AIs are meaningfully important for helping the AIs to improve at the top level.

I never said human games are meaningfully important for training AI. Human games are still important for the advancement of chess, maybe Magnus Carlson can learn from games between two Super AIs but most humans still learn from games by humans, Grandmasters are continuously developing the opening, middle-game and end-game systems, adding to the chess knowledge base. Every serious chess player still reviews and study games by prominent Grandmasters, every serious chess player documents their own games, writing down every move. All rated games are recorded and added to the chess database that every player can review and study.

>Does it? Or will AI be able to run its own experiments and find better/more efficient abstractions...

Only if it is in the training data.