I think continuous learning must modify the model because I don’t think infinite context is a tractable problem

So at some point you need to “bake” in accepted truths, ideally indexed to a formal memory structure for later validation

they definitely took all the disconcerting scenes from the movie iRobot and put them in this advertisement...

I do often think about how humans say tools become extensions of themselves, and what might this mean for an AI... it seems like AI needs a way to learn new modalities on the fly

it's not enough to have a "joint position" expert, it needs to be able to make/prune experts as needed

This is actually a good reminder that robotics is hard and what other groups are achieving is actually amazing

I view this kvetching as part of the process. We need these people churning up activists working for the right cause

As long as we have other people making gains in the right direction

I’m afraid to use the mini models, there’s a point they introduce more problems faster than they write code, and I’m better off just writing the code or paying for more accuracy