Well, that didn't take as long as I thought! I figured it'd be at least until the end of the month... :-)

(Short summary: smart Stanford researchers with a public-domain AI system used the super-duper corporate AIs to train their inexpensive model. They spent ~$500 training the Stanford AI this way, basically harvesting huge amounts of money and time spent on all the previous models for free. And they weren't even very efficient. Anybody who wants to can trivially have a pretty top-notch AI running on inexpensive hardware Any Day Now.)

https://crfm.stanford.edu/2023/03/13/alpaca.html