the biggest problem with predictive algorithms and LLMs is "truthfulness".

it is impossible to know the accuracy of the output unless it is a mathematical query or directly judged by a human

without the ability to audit, it will be relegated to "cute" and "unimportant" use cases