The case for guiding AI answers by how they should feel
One user argues that rules for getting AI assistants to answer correctly keep multiplying to cover edge cases, and proposes describing the intended feeling instead.
TLDR
Rather than keep adding rules, a user wants to describe what readers should feel when reading an AI assistant’s response. They argue that this carries more information per token than explicit rules, which they see as too sparse and requiring endless additions for edge cases.
Combined views
1.2K
1 Source, first seen ago