We need to stop evaluating AI based on what it is sold as by the bro-vestor.

It is not intelligence and never will be. It's an tools that when use in the correct context and under good guidance, is a fantastic productivity enhancer.

But it's not a Hal9000, or an Andromeda. Don't expect it to be a horse, when it is clearly a car.

I have not had an AI hallucinate on me in months. Because I always make sure it has a good brief and reference info and that the context window is not exceeded, and that if the context window does get exceeded, that it re-grounds itself on the reference information and asks for human guidance.

AI hallucinations are mostly operator error these days, either using it incorrectly, not giving it sufficient guidance or asking it to do things it's not suitable for.

For example, you would never want to use AI when your inputs and outputs have predictable criteria. You would pass those through a normal logic tree.

Where an AI would be suitable is trying to Triage situations where the logic gets an unhandled use case. It would evaluate the inputs and consider if any of the outputs sufficiently matched the inputs criteria. If it can't it marks it as a genuine exception. But a human still needs to give it the correct guiding criteria.