How do you help clients manage expectations about the potential and limitations of AI in their specific industry?

I usually think of AI as a high school student trainee, who has unlimited time to learn and read but has limited understanding of the quality they produce. Since AI does not yet have a general intelligence that would help it to determine the output's quality, it often produces mediocre or even bad output.

To manage clients' expectations I suggest the following:

- Be honest about the limitations from the first meeting.
- Show examples of inconsistencies (different output for the same prompts each time), potential errors (miscalculations), hallucinations, and misinterpretation of the instructions.

This will present an opportunity to speak about risks and how you can mitigate them together. This will definitely put you in a more trustworthy position compared to the others, who only speak about how everything can be solved with AI.