Artificial intelligence · 2017
AlphaGo Zero
Starting from only the rules, self-play surpasses every prior Go program in 40 days.
AlphaGo Zero learned with no human games, a single residual network and simpler search. It beat the Lee Sedol version 100-0 after three days of training. It showed human data could be a ceiling rather than a foundation.
What it built on 3
What it led to 1
Related measurements
The dashed line marks when it first worked.
Training compute of notable AI models Source: Epoch AI, Notable AI Models (CC BY 4.0)
ML hardware FLOP/s per dollar Source: Epoch AI, ML hardware (CC BY 4.0)