Four: Brian Christian on artificial intelligence
Effective Altruism: Ten Global Problems – 80000 Hours
Is Linearity in the Output a Problem?
In the mccullick and pitt's original version, it's kind of this zero. Later we have moved to continuous outputs and learned things like you need to have a non linearity in the output. So your output can either be sigmoidl or, more recently, this kind of rectified linear output. And people have continued to experiment with different activation functions. The key thing is that there's some non linearity. Beyond that, there's a little bit of it, sort of a little bit voodoo, in terms of which things appear to work.
00:00
Transcript
Play full episode
Remember Everything You Learn from Podcasts
Save insights instantly, chat with episodes, and build lasting knowledge - all powered by AI.
