@sapinker : Today's AI models, after being pre-trained on pretty much the entirety of the Web, and fine-tuned to give coherent responses to user questions, are subject to Reinforcement Learning from Human Feedback (somewhere BF Skinner is smiling). The problem is that the binary • TwiCopy

Steven Pinker

@sapinker

+ Follow

Cognitive scientist at Harvard.

ID: 107225267

linkhttp://www.stevenpinker.com calendar_today21-01-2010 22:20:32

13,13K Tweet

843,843K Followers

84 Following

Steven Pinker

@sapinker

2 months ago

Today's AI models, after being pre-trained on pretty much the entirety of the Web, and fine-tuned to give coherent responses to user questions, are subject to Reinforcement Learning from Human Feedback (somewhere BF Skinner is smiling). The problem is that the binary

thumb_up_off_alt152

chat_bubble_outline0

repeat39

shareShare