Jacob Steinhardt, UC Berkeley: Machine learning safety, alignment and measurement
Generally Intelligent
00:00
How to Learn Non-Adversarial Data Sets
You can still learn as if you had learned e non adversarial data set. You have to be willing to charge your model. There are these many image net models, and then like full scale image net models. But i think on all these models, the early layers look like edge detectors and color contrast detectors,. Sometimes the later layers just look kind of like mush.
Play episode from 16:18
Transcript
