Edit

I think the direction of these models seems clear, and their capabilities however limited at the moment are outstanding. I think the best measure of performance is what the naysayers complain about. We've quickly gone from "actual gibberish that looks like paragraphs" to "yes but it loses track after a few sentences" to "yes it's kind of coherent but doesn't make sense totally" and are now at "it's like an overconfident human at many tasks, when they're only somewhat qualified in the area". People are pointing out where it's got an API call wrong, without it having any ability to look up the answer and see if the function actually exists.

I spoke to it about distributed computation design and discovered something new I can actually implement in my design.

I'm very excited about where this will all go because it seems useful now and I'd be very surprised if this is the best that it ever reaches.