I would be really interested to understand the efficiency of this learning vs a more traditional approach. If you have two models - ChessGPT and a traditional chess-playing ML model - and train them on the same number of games, which reaches a higher level of chess-playing skill?