I've been playing with with ChatGPT and Gemini and other AI robotic overlords for the AI chapter in the 4'th edition of Tcl/Tk: A Developer's Guide. I gave Chat GPT a fairly simple math word-problem and got this answer:

....................................................... Tokens used: 176 OpenAI answers {screws cost .10 each} {use 1 screw per 3 feet} {nails cost .05 each} {use 1 nail every foot} {what is the most economical way to fasten 12 feet of siding} with

In this case, using nails would be the most economical way to fasten the 12 feet of siding.

For the screws: - Cost per foot = $0.10 / 3 = $0.0333 - Total cost for 12 feet = $0.0333 * 12 = $0.40

For the nails: - Cost per foot = $0.05 - Total cost for 12 feet = $0.05 * 12 = $0.60

Therefore, using nails would be cheaper at a total cost of $0.60 compared to $0.40 for screws. .......................................................

For what it's worth, the Gemini AI from Google got that answer correct, but when asked who wrote my book, gave me the authors for a book I didn't write.

One of my computer courses in college finished up with the professor explaining all the ways you could get a bad answer from the stuff we'd spent a semester learning. "The one thing about using the computer is that people will trust the answer if the computer gives it. If it's printed on GreenBar paper, they think it's true."

I am not afraid of our robot overlords taking over. I am deathly afraid of the people who trust answers they get from an AI.