What if instead of talking about “AI” we talk about “automation”?

Big mistake.

One defining characteristic of currently fashionable LLMs is that they are only right 70-80% of the time, which works fine for "chatbots" that are directly in communication with people, they are completely untenable for "automation" tasks that take multiple actions on their own. (e.g. put 10 80% reliable steps in a row and you have a process that works only 10% of the time)

That is, you can ask ChatGPT, like Alexa, to "open the pod bay doors" but on a flight to Jupiter they'll struggle to get out of LEO.

Why not Behavioral Intelligence? After all that’s what we really mean. We want things to “behave” in certain ways, not necessarily automatically, rather responsively.