Accuracy and AI limits

Why it happens

Iris runs on large AI models. They are very good at producing fluent, plausible language, which is exactly what makes their mistakes easy to miss. An answer can read as confident and still be incorrect, out of date, or missing context.

This is often called hallucination. It is a limit of the underlying technology, not something any AI product has fully solved.

Where it shows up

• Recent events. The models have a training cutoff, so they can be confused, or simply wrong, about things that happened recently, unless Iris searches the web for them.
• Quotes, numbers and citations. It can produce a statistic, a quote or a reference that sounds authoritative but isn’t grounded in a real source.
• Niche or unfamiliar topics. Where it knows little, it can fill the gap with something plausible instead of saying it doesn’t know.
• Your own context. Memory helps, but Iris can misremember, apply an old note to a new situation, or miss something you never told it.

How to check

Treat Iris as one input among several, not a single source of truth.

• When Iris searches the web or reads your files, open the sources it cites rather than relying on the summary alone. The original may carry context the answer left out.
• If an answer depends on something it remembered about you, check the note. You can correct or delete it in your memory.
• Ask Iris how sure it is, or what would change its answer. It can usually tell you where it is on shaky ground.

When it matters most

Double-check anything you plan to act on, and take particular care with medical, legal, financial and safety questions. Iris is not a doctor, lawyer or financial adviser, and its answers are not professional advice.

If you are working with information that matters, verify it independently before you rely on it.

Tell us when it’s wrong

Every report helps us make Iris better. If it gets something wrong, tell us what you asked and what it said at iris@ferrolo.com.