yeah, it's still developing at a breakneck pace and there could be huge changes just over the horizon, it's hard to know.
my other interest is in just how useful small models can be made, which can run on constrained hardware. sort of the other end of the "how can I get a really good GPU that fits in a 1LPC in my closet" problem - how can I get a really good model that runs on the 5yo 1LPC already in my closet.