Google DeepMind paper trains an agent to rewrite its own harness

3 points | by runtimewire 5 hours ago

No comments yet.