AGI, the holy grail of AI development, masters any cognitive task a human can do, learns across domains, and improves itself. And the core building blocks are already here.

But when systems self-improve, even 1% misalignment = P(doom) 💀

Let's make sure we get it right.

🧵