Why AI models hallucinate and which models get it wrong most
KARE 11 | September 23, 2026
In this KARE 11 report by Chris Hrapsky examining why generative AI models hallucinate and how their accuracy varies, VILAS co-founders James Holmberg and Sarah Anderson Soubra discuss how people can use AI more effectively while recognizing its limitations. James explains why working with AI requires practice, specific instructions, and meaningful context. He also shares practical ways to reduce errors, including verifying sources, refining responses through conversation, and applying human judgment rather than accepting AI-generated information at face value.
How to use AI without blindly trusting it
If AI is a mistake-laden, bootlicking genius, how do you use it to your advantage?
"It is almost infinite what it can do," said James Holmberg, co-founder of VILAS AI.
Holmberg and co-founder Sarah Anderson Soubra teach companies how to efficiently and responsibly use AI while reducing the risk of hallucinations.
"It is not one size fits all. We think of it as an instrument. So you have to practice and every person will play it a little differently," said Holmberg.
Their tips? First, prompt specifically. Tell the model not to guess or infer anything. Second, ask for links and sources that you can double-check.
"Even go a step further and ask it to explain how it came to this answer," said Holmberg.
Third, stay focused on one topic per thread. This keeps the model's focus narrowed and more accurate. And finally, rarely accept the first answer. Refine your task through conversation with the model.
Holmberg compares AI to an employee:
"This is your intern sitting right next to you. It is a great intern. You have to delegate things, put in the context. But just like any intern you have you are going to review everything. You are not just going to say ‘oh this looks good’ and pass it off."