10 ms·
About twenty-five years ago there was a web site that claimed a better Turing test was to ask a large number of relatively easy yes/no questions, with the idea
by gcanyon 2y ago
About twenty-five years ago there was a web site that claimed a better Turing test was to ask a large number of relatively easy yes/no questions, with the idea that a human could answer virtually all of them correctly. Toward the end of passing that test they were collecting binary questions with their answers, e.g.
Is the Moon made of cheese?
Is the sky blue?
Has there ever been a human being twenty feet tall?
As far as I know it disappeared a long time ago — I’ve done lightweight research and not found it.
It’s too bad, it seems they were onto something.
- sebzim4500 2y agoI feel like GPT-4o would ace that. If you look at the questions in this benchmark they are stuff that 99% of people won't know, although at least they would know that they don't know the answer.
- gcanyon 2y agoWe'll find out -- I just submitted a form to HN to collect some QA pairs to try it out.
- antiquark 2y agoSounds like the computer game "Animal" from 1978. https://www.atariarchives.org/basicgames/showpage.php?page=4 https://www.atariarchives.org/basicgames/showpage.php?page=4