I agree. I think the next big advancements will come from a different model, not from fine-tuning current ML setups.

I'm more interested in a general AI that can learn any game or environment it encounters to optimize a return. Not quite general AI, but a different path than what is going on right now.