@rowat_c AI Overview (just triggered by search in the browser with Google)

honestly this one feels kind of right

"While less likely between a husband and wife, it's possible there's something more than friendship involved."

a useful tip as you are getting ready to go out this weekend

please don't try at home

Apparently the point of the Tortoise and the Hare was simply that turtles are just awesome distance runners.

Sometimes you have to wonder if AI Overview is just too deep for us mere mortals...

ghosts part 2 -- I figured this model wouldn't go astray on the ghosts example from before, but in fact it produced some... interesting advice (that I have yet to try).

dressing up as a ghost (part 1)

logic

It's always instruction #3 isn't it.

"That's my answer and I'm sticking to it."

Often, asking ChatGPT (o3) to draw something in Beamer (tikz) gives better results. This staircase made out of triangles has some drawbacks though.

the AI perspective on carbon taxes

"I'll let my kitchen AI choose its own energy source, what's the worst that could happen..."

AI Overview doubles down

(Thanks to my colleague Ryan O’Donnell for asking for a direct visual proof of this.)

Enjoyed taking part in the millennia-old tradition of visual proofs in geometry, with this upcoming note in The American Mathematical Monthly. The picture proves the parallelogram law. Do you think AI can find such proofs yet? Paper: https://www.cs.cmu.edu/~conitzer/visualAMM.pdf

Nice to see Peter Stone featured in People of ACM, also talking about AI100 and our corresponding recent Communications of the ACM article (with Eric Horvitz and Sheila McIlraith).
https://www.acm.org/articles/people-of-acm/2025/peter-stone

Are the "reasoning" models especially vulnerable to this?