Laura Weidinger: Ethical Risks, Harms, and Alignment of Large Language Models
The Gradient: Perspectives on AI
00:00
Mispecification in Language Models
In linguistics, there are some examples of sentences that syntactically make perfect sense, but semantically are absolutely pointless. So what you could say has happened here is that actually you forgot to specify a kind of situation or context where you would want the AI to not just continue based on the prompt. And so these are just some examples of how mispecification can happen in language models as we were talking about in this paper.
Transcript
Play full episode
