
Episode 18: Oleh Rybkin, UPenn, on exploration and planning with world models
Generally Intelligent
00:00
The Old Radio Isn't Working, Is It?
The theory in this case comes rom from bagging, from the worsabushap ensamples write the bushop predictors. Mathematically, if you train them on different data, will give you some variance that is actually due tho variance of your estimater. There's also some other mechanism that gives you variants,. So maybe we just don't know what the mechanism is. Would be really interesting to find out. But practically, the old radio works. And i think i've removed the option from the code. I will probably thatsome point in the future. Maybe once it helps, i well start using it. If it works, it works. Probably. No theory yet that explains
Transcript
Play full episode