Why A.I. Says the Darndest Things
What Next: TBD | Tech, power, and the future
00:00
Security Guardrails on Large Language Models
The problem with implementing safety guardrails on large language models is that everyone, including the people making the systems, aren't entirely sure how they work. This isn't like traditional coding where you type some commands in a certain programming language and the computer responds in a predictable way. They have been taught not by specific instructions, but by reading the Internet,. The answers are so interesting, but it's also just a really different way of human machine interaction.
Play episode from 18:19
Transcript
