We may not need smarter AI. Just less stupid AI. The big problem with LLMs is that most of the time they act smart, and some of the time they do really, really

We may not need smarter AI. Just less stupid AI. The big problem with LLMs is that most of the time they act smart, and some of the time they do really, really dumb things and don't notice. It's not the ceiling that's the problem. It's the floor. Which is why, as the article points out, "agents" aren't very useful yet. You can't trust them to not screw up big-time.