Recently a friend of mine has begun learning guitar. He’s a smart guy and he’s picking it up pretty quickly, but he’s vibecoding some learning tools using claude and claude doesn’t know how to play guitar.

He generated some visualizations of the fretboard, but each fret was numbered (starting at A) 1, 1.5, 2, 3, 3.5, 4, 4.5, 5, 6, 6.5, 7, 7.5. If you know a bit about scales, you probably could recognize this pattern as an A minor scale. My friend, being new to music, did not know that. He just knew that he now had numbers to correlate to posistions on the fretboard.

When I asked what scale it was showing him, he said the guitar was in standard tuning. Of course, he wasn’t sure what scale it was. He’s brand new to music! The generated app wasn’t helping him learn and was making it harder for him to communicate to others, since he was learning from a unique, detached source.

This really highlights a big problem with relying on AI for self-learning. You don’t know what you don’t know and neither does the AI. If you don’t know what to ask, the AI won’t know what to make for you. If you’re not knowledgeable, you can’t be sure whatever is generated is correct or useful.

This is why “AI psychosis” has become such a common term to refer to people’s tendancy to belive AI until they’re so far removed from others that it’s hard to communicate about the same subject.