Why We Write and Language Models with Maggie Appleton
Clerestory (Bryan Kam)
00:00
The Dishonesty of Machine Learning
Machine learning can't trust words that are written but if none of the words we read have any anchoring in like grounded experience of individuals then they have no anchoring in truth or what actually exists right. Everyone's just like talking to each other on you know in an abstract land without actually checking that stuff's true or real or yeah and that holds regardless of what your view of truth actually is so we were talking on sunday about correspondence theory of truth where we try to get our concepts to match quote unquote reality. We think it's a great idea to let language models write a lot of this text without any way for them to check that external reality yes unless sure okay we want to
Transcript
Play full episode
