Simple enough question, one that you would think any AI model should be able to answer. I mean you don’t need to ingest a huge amount of data to train on for this. Yet, as you may have already seen, a school in Kentucky discovered that you can’t trust all of AI’s output.

The start of the academic year booklet, sent out to all students and parents, was riddled with errors, including this howling representation of the states that make up this fine country. Leaving aside the lack of review from any upstanding adult for the moment, surely this has to be coming from an older AI model. Or is it? Let’s see.

Google’s Gemini attempting the same

Let us see if the latest Gemini model from Google can do any better. I gave it this simple prompt:

Create a map of the USA, show all of the states. I want it in color

Off it went, and it produced this as its first attempt:

At least it had the honesty and self awareness that it got most of it wrong (though lets celebrate it named the Gulf of Mexico correctly). Even this Scotsman could spot something wasn’t quite right, and asked it to expand all the state names and be consistent.

The following was produced:

Getting better at the state names being correct. Even though I didn’t ask, it attempted to call out the state capitals. Google got a little sniffy when I pointed out its errors, doubling down on its accuracy.

Mmmm I think not. Thought I would nudge it in the right direction and call out a couple, to which it corrected the map to:

I still pushed it for another correction:

At this point I gave up, even though I was on the most advanced model of Google at the time.

AI can make mistakes

This is a fascinating experiment, as it highlights how far we still have to go with AI models. Rendering a list of US states is not something you would be expecting to need to utilize the latest and most expensive frontier models for. Yet here we are – August 2026 and Google is failing to render a map of the United States at a reliable manner even after rounds of corrections.

It would be easy to say that someone needs to be fired in Farley Middle School for not checking this literal AI slop before it was sent out to its students/parents, but how often is AI BOVINE (Being Overly Verbose In Needless Expositions) truly vetted within the corporate environment? How often is all AI code actually reviewed and checked?

We can all have a good giggle at the mistake of Illinois becoming “Vitoiis” because it is easy to spot. Other mistakes are not as obvious and can sneak through unchecked through many eyes.

If there is any takeaway from this, is to stop blindly relying on AI’s output, no matter how confident it sounds, or assures it is correct. It will get it wrong and does argue with you before admitting its fault.

AI is a tool, not a thinking replacement.