It’s not about truth, it’s about acceptance. That’s why novels, myths and stories are ‘acceptable’ even when they have no basis in reality. The make the reader feel good.
The same applies to LLMs. I’d argue that most users don’t actually care about reality- they just want something acceptable and to feel good. The trick is to make sure that they know it’s actually fiction or adjusted reality.
The equating is not one to one, but becomes clearer when described:
- novels: imaginary examples to force specific situations and postulate on the outcomes
- myths and stories: patterns of reality that apply at levels at or above a single human life, exemplified in a language that can be understood instinctively if not rationally
- llms: responses that tend to the average cluster of information related to that concept as described in the training data
Nice. Thanks!
