Simple Social Network

{"p":"","h":{"iv":"ROXSYW+cfvEbFHu5","at":"ocxplSQjdRC3tXEtB/9/wg=="}}

Darius Kazemi

@trochee here's the link for your reference. It's pointing to some jargony stuff about model calibration

Jeremy Kahn

@darius that's a lot of words and jargon to say "imitation learners shouldn't extrapolate, but they do"

…all while missing Emily (&c)'s point that the models don't actually "know" *anything* , they just produce statistically probably sequences