Self-Play
Improving by training against copies of oneself.
In self-play, an agent learns by repeatedly competing against versions of itself, generating its own ever-harder curriculum. It is how AlphaZero reached superhuman play in games with no human data.