from Hacker News

We need a new “Turing Test”

by tvvocold on 3/29/23, 2:50 AM with 3 comments

• by skybrian on 3/29/23, 3:34 AM

  When people informally say something "beat the Turing Test," that has little to do with Turing's proposed imitation game, which is more of an interactive party game like Werewolf.

  It's kind of a weird benchmark. I think it's better to think of it as just another challenging game for AI researchers to try to beat, like chess or Go or Diplomacy. Like those other games, how tough it is to beat will depend on the human players. I think a proper test would be to play against skilled human players who have practiced.

  I wrote more here: https://skybrian.substack.com/p/done-right-a-turing-test-is-...

• by akasakahakada on 3/29/23, 3:58 AM

  The demonstration here is not working. The author should ask GPT-4 to provide a piece of code or so to disprove the convergence of 3n+1 problem, instead of in form of the usual vague unidirectional math exercise.

  Sometimes usage of the user reflects the user's ability. It is more likely a test to find intelligent human.

• by abudabi123 on 3/29/23, 2:58 AM

  Use of current buzz-hype A.I. to infer from that original 50s paper will go so far.