new account, logged out of my anon.twitter.com/dylantirandazgithub.com/dylantirandaz
currently building a biocomputing startup, moving to sf.
This seems very interesting, i actually think the lack of humor in a model is more of a mode collapse issue. Being able to verify humor would be nice but isn't this something that would be very prone to reward hacking?
This seems very interesting, i actually think the lack of humor in a model is more of a mode collapse issue. Being able to verify humor would be nice but isn't this something that would be very prone to reward hacking?