#49 - AGI: Could The End Be Nigh? (With Rosie Campbell)
Increments
00:00
The Pros and Cons of Reinforcement Learning With Unlimited Data
I think I, yeah, I am pretty concerned that that could get us to. This is the systems that would be very dangerous. If we march towards this goal there's just really not going to be anything we can do to avoid the harm or the catastrophe. And then there's some amount of probability on like. We're marching towards that, but we are putting effort into the right places in terms of interpretability research and other safety engineering approaches That will constrain it in a certain way. All of that stuff is going to maybe get us on the right track to avoid.
Transcript
Play full episode
