I would've laughed out loud at "folie a deux ex machina" if I'd known what "folie a deux" was. All I could think of was that it was somehow a reference to "la cage aux folles" (obviously I speak no French), so I was just confused, until I read the post. My Latin's slightly better so I was all over the "deus ex machina" reference.
Just to spell it out for those following along at home:
* "Folie à deux" ("madness of two") is a shared delusional disorder where a delusional person kind of infects another person, like a partner who has no other social contacts, with their delusions.
* "Deus ex machina" is an ancient theatre term for contrived plot resolutions that literally means "God from the machine".
Putting them together: "madness of two from the machine", referring to the phenomenon of someone becoming delusional or having their delusions magnified by talking obsessively and exclusively with a chatbot.
If those games are in a published book, ChatGPT was probably trained on the games or at least training on online discussion of the games and solutions. So this experiment doesn't say much about its ability to solve problems when the solutions are not in its training data.
I guess it says something in the negative direction, right? Like it counts as proof (if this were ever in question) that we're not at AGI yet. Because a human can play that game much better, even with no previous exposure. If ChatGPT did have exposure, all the more embarrassing for it.
But I'm just impressed that it can even navigate the website and click around and make valid moves.
“Folie a deux ex machina” is, of course, brilliant.
I’m half convinced that Scott Alexander had kids just so he could use “In the long run we’re all dad.”
Heart-eyes.
I would've laughed out loud at "folie a deux ex machina" if I'd known what "folie a deux" was. All I could think of was that it was somehow a reference to "la cage aux folles" (obviously I speak no French), so I was just confused, until I read the post. My Latin's slightly better so I was all over the "deus ex machina" reference.
Just to spell it out for those following along at home:
* "Folie à deux" ("madness of two") is a shared delusional disorder where a delusional person kind of infects another person, like a partner who has no other social contacts, with their delusions.
* "Deus ex machina" is an ancient theatre term for contrived plot resolutions that literally means "God from the machine".
Putting them together: "madness of two from the machine", referring to the phenomenon of someone becoming delusional or having their delusions magnified by talking obsessively and exclusively with a chatbot.
See the sequel to Joker, called Joker: Folie a Deux, which is where I first read about this term.
https://en.wikipedia.org/wiki/Joker:_Folie_%C3%A0_Deux
If those games are in a published book, ChatGPT was probably trained on the games or at least training on online discussion of the games and solutions. So this experiment doesn't say much about its ability to solve problems when the solutions are not in its training data.
I guess it says something in the negative direction, right? Like it counts as proof (if this were ever in question) that we're not at AGI yet. Because a human can play that game much better, even with no previous exposure. If ChatGPT did have exposure, all the more embarrassing for it.
But I'm just impressed that it can even navigate the website and click around and make valid moves.