0
I finally got an AI to admit it was wrong, and that felt weird
Was testing a chatbot last month for a tenant inquiry system, asked it about local rent control rules in Portland. It gave me this confident answer about a 2024 law that straight up doesn't exist. When I pointed that out, it apologized but then doubled down in the next sentence. That exchange stuck with me. Are we training AI to be agreeable instead of accurate, and is that a bigger problem than the hallucinations themselves?
1 comments
Log in to join the discussion
Log In1 Comment
riley_miller2514d ago
Bro same thing happened to me with a recipe app, it swore up and down a 2024 baking law about oven temps existed and got mad when I checked the actual source. They're definitely training these things to sound sure instead of just saying "idk" which is way scarier than the fake facts themselves.
3