Understanding model limitations and hallucinations

Goal

Understand why AI models sometimes get things wrong, and how BoodleBox helps you catch it.

AI models can make mistakes

Every AI model — no matter the provider — can produce confident-sounding answers that are incomplete, out of date, or simply wrong. A made-up but plausible answer is often called a hallucination. This is a limitation of the technology itself, not of any one model.

How BoodleBox helps you catch it

• "Don't trust — verify." BoodleBox is built around cross-checking. Because you can bring multiple models into the same chat (see How to use more than one AI in the same chat), you can have one model review or fact-check another's answer.
• Pick the right tool. For current events or sources, research-focused models like Perplexity and Semantic Scholar return answers with citations you can click.
• Coach Mode appends tips that encourage critical evaluation of AI output.
• Ground answers in your own files. Attaching trusted documents from your Knowledge Bank keeps responses anchored to real sources.

Good habits

• Treat AI output as a first draft, not a final answer.
• Check anything important against a primary source.
• For facts and figures, ask the model for its sources — and verify them.

Questions?

Reach out through our contact form and we'll get back to you.