A tsumego solver using deep Q-learning. https://en.wikipedia.org/wiki/Tsumego
The three unmarked white stones are the initial stones. Black wins if he captures all white stones. Black 1 is a natural move, trying to destroy white's eye. White tries to expand his eyespace by playing white 2. Black 3 looks like a bad move, but white ignores it and creates an eye with white 4. This lets black play black 5 to save black 1. Black 3 now looks like a well-placed stone.
All moves in this diagram are bad lol. White 1 is probably trying to stop black from falsifying white's eye. Black 2 looks like it might be trying to threaten to falsify white's eye. White 3 is a terrible move, filling white's own eye.
Black 1 looks kind of reasonable. White tries to make eyespace in the center with white 2, but black 3 is the vital point, removing any eyespace there. White captures the two black stones with white 4, good.
Though it looked like white had two eyes, black 1 is a great move known as a throw-in, the only way to kill white in this position. White is now completely dead. The game continued until all white stones were captured and black won.