Source for what ? Do you have any idea of what you're talking about ? Hold on, How do you think Language models work ? What do you think the plain English Instructions are for ?
Re:AI knows things no one told it
by hammyhavoc on Friday May 12, 2023 @10:07PM
(#35924766)
A source for how guardrails function as what you state isn't what's documented. You said that the LLM chooses when to use them.
The LLM sits behind the toolkit which sits behind the guardrails. Are you sure you understand how they work?
The guardrails monitor your input, the output of the LLM, and modify the output to suit, including terminating the chat.
It should be in line with the likes of this - https://arxiv.org/abs/2302.04761 or https://twitter.com/minosvasilias/status/1627076214639976449
I didn't say guardrail to mean anything particularly elaborate. The pre-prompt for Bing has instructions to avoid adversarial arguments.
"RAIL (Reliable AI Markup Language) is a language-agnostic, human-readable format for specifying structure, type information, validators, and corrective actions for LLM outputs. RAIL, an XML flavor, allows users to define the expected structure and types of LLM outputs, the quality criteria for valid output, and the corrective actions to take if the output is invalid."
Why do you keep projecting your own humanity onto the LLM?