aiExpert FAQ

Q: What is aiExpert?

A: aiExpert is part of Model Expert, which is in turn part of the eaTeamWorks family of products.

Q: Does aiExpert cost more?

A: No. aiExpert is included in Model Expert, for the same price as before

Q: What will aiExpert do with my confidential information?

A: aiExpert does not store your data, and the LLMs which process it don’t either. So nobody sees your data except you. See Data Confidentiality

Q: Will aiExpert make changes to my model?

A: No. aiExpert will make suggestions about how you might improve your model. It’s up to you to follow the suggestions or not.

Q: What about hallucination? Might aiExpert just ‘make things up’?

A: This was a real problem with early AI models. It’s much less of an issue now. We have not seen any hallucinations in our testing, but keep a look out.

Q: What about consistency? Will I get the same answer to the same question?

A: Not really. Asking an AI to make suggestions about your model is a lot like asking a human expert. Mostly, they will give approximately the same answer to the same question, but there will be random variations.

Q: Can aiExpert deal with our local standards?

A: Yes – you are encouraged to adapt the Model Query prompts to add in your own local rules and standards.

Q: Will aiExpert understand the model domain we are working in?

A: Usually, yes. And we were surprised about this. Because the AI has such a huge amount of knowledge of the world, it can give amazingly accurate advice, on some domains.

Q: Can i run aiExpert against my whole model?

A: No. aiExpert only runs against single diagrams, and we know that the more data we send it, the more confused it gets. So sending it the contents of a diagram with up to 30 ‘things’ on it seems to return sensible answers. More than that, and the quality of the feedback falls away. Little and often.

Q: My users are not first-language English speakers. Can it help them?

A: Yes. Each Model Query specifies the national language in which the modeler wants the response. This is a standard feature of LLMs, but we have not had the chance to test this in detail. If you use this, please send us some feedback.