It made me realise my new coworker is an moron and a fraud in just weeks, not months or years. If he had simply shut his mouth, I would never has suspected him. But thanks to him feeding several internal things into AI, and then bragging about it to my boss, I now know who I won't be sharing all but the most required info with.

It also amuses me when it still fails at some basic tasks. Having several AI models fail in different ways is quite interesting, so there is some entertainment value. Did you know that Deepseek and Qwen do not know the lyrics to "Gansta's paradise", while Claude knows them if you poke it a bit, but runs into its profanity filter?