My AI assistant totally failed me during a big client pitch

Last Thursday, I was showing a new AI tool to a client in Chicago that's supposed to write marketing copy. I gave it a simple prompt for a product description, and it spit out complete nonsense with random numbers and symbols. The client just stared at me for like 10 seconds before asking if it was broken. I had to scramble to explain it was a 'training issue' and switch to a demo I'd prepared earlier. It made the whole field of generative AI look super unreliable in that moment. Has anyone else had a public demo go wrong because the model just decided to act weird?

3 comments

3 Comments

king.derek13d ago

I had a demo in Austin where the AI started repeating the same three words over and over. My rule now is to have three pre-generated outputs saved in a doc before I even walk into the room. That way, if the live model acts up, I can just paste from my backup file and say we're showing a different style. I also run the exact prompt I plan to use at least five times from the same coffee shop Wi-Fi I'll use on-site to check for weirdness.

the_rose13d ago

Smart move testing on the same wifi, that's next level prep. I'd also save screenshots of the good runs to prove it worked before. Maybe even film a quick screen recording as a last resort backup. Honestly your method turns a potential disaster into a planned demo of how you handle edge cases.

xena_fox3921h ago

Five test runs on the same wifi is the absolute minimum I do now. Honestly, after seeing what happened to @king.derek, I started keeping a hidden text file on my desktop with perfect outputs ready to paste. Tbh, filming the screen is a solid last resort, but I also make sure to have the exact demo script typed out in a notepad doc just in case the audio fails. That way you can just read it live and no one knows the difference. It turns a total fail into you just calmly switching to a backup plan.