That time I trusted a chatbot's answer and it cost me 3 hours and $60
I was at my desk in Portland last Tuesday trying to get an AI tool to summarize a long lease document for a client move. The model was confident, said the clause about building access was on page 12, and I repeated it to the property manager without double checking. She read it back to me and it was nowhere near that page, so I had to call the client and apologize, then dig through 40 pages myself. I still use these tools, but I think the hype around them answering anything is flat wrong, and I want to hear from folks who have found a better way to check their output before it leaves your hands.
That part about page 12 blows my mind. I've had models do the same thing where they invent a page number with total confidence and it's a 40 page doc, so the odds of it being right are almost zero. A friend of mine works at a law office here in Portland and they print the whole lease and highlight the real clause before anyone repeats it out loud. Reading it back to yourself before you send it out is the only way I've found to not eat that kind of mistake.