If I told you one of the most harrowing looks at how dangerous LLMs can be to the human psyche came from a YouTube video originally intended as a comedy, would you believe me? The danger of having a digital sycophant willing to believe and affirm everything you tell it is backed up by any number of real-world violent incidents. For anybody struggling with mental illness, ChatGPT is their worst enemy disguised as their best friend.
While we hear horror stories of AI-fueled delusion with increasing frequency, mustachioed investigative YouTuber Eddy Burback decided to deliberately create one to get better insight into this process and how it happens. Starting with a single intentionally comedic belief — "I was the smartest baby in the world in the year 1996" — was enough to kick off an intense downward spiral, with ChatGPT using that one statement as a launching point to get Eddy to abandon his family, move to an RV in the middle of nowhere, tap into radio towers to amplify his own brainwaves, and more.
Even a relatively mentally sound person aware of the difference between reality and fiction was deeply affected, in large part unintentionally. For those who aren't so lucky? It's no wonder LLMs like this are having such widespread effects on those who are most vulnerable. It's a hard hour to get through, but it's absolutely essential to understand just how insidious these systems are.
