The humans were prompted to read the AI responses very carefully because their hallucinations are very good at convincing you with words. It takes a certain ski

The humans were prompted to read the AI responses very carefully because their hallucinations are very good at convincing you with words. It takes a certain skillset to question every word that comes out of a language model because most people will go “hmm yeah that logic seems right”. So hiring “smart” people is insufficient, you need very paranoid people who question every assumption.