3 ms·
The point -- which I don't think you got -- is that extremely generic ingredients like high-quality data (which is the point of Stockfish here) and very deep Tr
by ogogmad 3y ago
The point -- which I don't think you got -- is that extremely generic ingredients like high-quality data (which is the point of Stockfish here) and very deep Transformer-type Neural Networks, are enough to nearly match the performance of ad-hoc, non-generalisable techniques like gametree search algorithms.
This has two possible applications: 1. There's far less need to invent techniques like MCTS in the first place. 2. A single AI might be able to play grandmaster level chess by accident.
The catch is you need high quality data in large amounts.
- YeGoblynQueenne 3y agoI did get the point and I'm commenting that the point is missing the point. There is nothing new in learning that a large neural net can approximate the output of a classical system. This has been done many times before. The real point is that DeepMind build a system that is half-search and pretend it's no-search. You cannot get the "high-quality data" without a classical system- not in chess.
- ogogmad 3y agoI get your point. Acquiring the data is the hard part, and they cheated to get it. It's chicken and egg indeed.