Hacker News Reader

Comment by civilized | Hacker News Reader

You'll prefer entertaining AI when you're in the mood for entertainment. What about when you just want something done so you can move on with your life? Will you appreciate a robot playing fake emotional mind games?

View on HN

Yes, that's what I was saying when I was noting it would probably be not as fun when I'd have to encounter it in real life. There's some analogy to some other areas where things that are terrible are hilarious or extremely entertaining until one is thrust in that scenario, and then it is simply cruel irony (as best originally defined in "The Emperor's New Groove", IMO).

This reminds me of the Verizon debacle where the user couldn't convince the customer service reps that 0.002 cents is not the same as 0.002 dollars. It was hilarious to hear the phone calls, but would have been maddening to experience. Arguing with an AI about the year is funny, until it's you arguing with it after you were billed for an extra 10 months of service.

The details of this debacle are still available at https://verizonmath.blogspot.com/!

Dial to the humor setting to 60% TARS

Presumably it either already does or can be made to have the equivalent of the linguistic concept of register.

“Acting as a customer service agent, enquire about how you can assist a customer” kind of prompt. There should be plenty of training data for how customer service agents interact and the kinds of words they use and attitudes they affect, and easy fine tuning for what is and is not appropriate.

This kind of statistical context seems like exactly what these language models are good at: sounding like what is expected. You lean on your customer service robot to interact a certain way and it will.

Reply on news.ycombinator.com