Yejin Choi: Teaching Machines Common Sense and Morality
The Gradient: Perspectives on AI
00:00
Taking a Pre Trained Language Model and Filtering Out the Bad Information?
We have these really powerful methods where we can combine nearat works with knowledge grasps, but that doesn't scale if knowledge grass has to be human generated. So why don't we take a massive pre trained language model like gpt three and filter out the bad stuff get to the good stuff? I'm wondering if you can expand a little bit on what that method for taking out the bad information looks like.
Transcript
Play full episode
