What is AlphaGo?

Eric Jang on how LLM coding has collapsed the cost of frontier AI research replication.

“Thanks to LLM coding, what took a whole team of research scientists at DeepMind and millions of dollars of research and compute can now be done for a few thousand dollars of rented compute.”

AlphaGo has already exceeded human Go-playing ability so far that its further progress is no longer interesting to humans because it is not learnable.

“We're already so far beyond human go playing capability that it's not interesting to humans because it's not learnable.”

Move 37 was innovative but not creative, contradicting the widely held view that it represented a creative breakthrough by AlphaGo.

“I don't think that move 37 was creative. I think that move 37 was innovative without being creative.”

AlphaGo's training techniques did not directly inform AlphaFold, contradicting the common narrative that AlphaGo led to AlphaFold.

“It's not like the Alph Go agent or indeed the Alph Go training techniques really informed Alpha Fold in a very direct way.”

Hybrid chess engines pairing a fast-updating neural network with classical search outperform pure AlphaGo-style neural engines in open-source competition.

“They took the neural networks and they made these like a certain type of neural network that can update really fast when you change the state and then they combine it with like just super fast search and it actually outperforms the best like open source like alpha go type chess engines.”