Neither AlphaGo nor AlphaZero were transformers. Why would you expect the same results?

Going further, current LLMs have at least an order of magnitude more computing resources, but completely suck at go. Why would this suddenly change unless we dumped the countless games played by alphago for them to train on?