
Episode 01: Kelvin Guu, Google AI, on language models & overlooked research problems
Generally Intelligent
00:00
Is There a Difference Between Positive and Negative Samples?
If you're always just sampling close the boundary, sometimes the model is just catastrophically bad away from the boundary. Another issue that seems to come up a lot with these contrasted things is that the true label in that setting often has some spurious correlation that lets you pick it out easily. There are only so many things that grammatically fit in a slot. And so you haveto, when you cut, you have to cut out a proper sentence or some other cture that reduces the correlation on the boundary. Some dogs just do look like cats, and you shouldn't really necessarily be labelling them one way or the other.
Transcript
Play full episode