[ExI] reverse turing test (kinda)

[spike at rainier66.com]

Hey cool, a contest.

 

A version of the Turing test suggested that if a human and a computer were
each given a question or prompt, humans would not be able to distinguish
which intelligence wrote which response.

 

OK, ChatGPT can write responses to prompts, but it is easy enough to tell if
the answer is machine derived.  We have software to detect when essays were
written by ChatGPT, which is how that college was busted for having ChatGPT
write a letter to the students after a campus shooting.

 

It isn't perfect, but that imperfection gives me an idea.  

 

We could have a group of human competitors intentionally write in the kinda
formal stilted way ChatGPT writes.  Then a group of judges would look at the
responses, exactly one of which is GPT generated, and try to figure out
which one that is.  We could have arbitrarily many competitors and
arbitrarily many judges.  

 

A human competitor gets a point for each time they gets identified as the
GPT.  The GPT gets a point for each time it is identified as a human.  In
each round, there is no score if all the judges correctly identify the
software-generated response.  If the competitors fool all the judges then
there are as many points awarded that round as there are judges.

 

I think I can write in GPT.  Can you?  Are you better at GPT than I am?  Who
is the champion GPT-er?  Is it GPT?  Or are humans better at GPT-ing now?

 

spike

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.extropy.org/pipermail/extropy-chat/attachments/20230611/07dd74bd/attachment.htm>