You can take any video you yourself made and voice translate it for free already in minutes. It's quite simple, but it will make some mistakes. Enough so that it's not quite useful without editing. Try your own video and see how it works.
Well that was an interesting experience… I’ll admit I didn’t even understand more than half of those questions. I now feel rather sorry for the poor AIs being subjected to those tests on a regular basis!
At least one of the questions on the test had at least two valid answers, as the question itself did not sufficiently define the pattern (that is, there were at least two patterns which explained the behavior equally well, which resulted in two distinct answers).
A more general issue I had with the quiz is that it wasn't clear whether or not information was allowed to overlap - for example, two line segments occupying the same space, which I'm reasonably certain did happen in at least one question, and which potentially renders -all- of the questions unanswerable even in principle (as it introduces an undefined number of additional invisible parameters which can modify the solution-space in arbitrary ways).
At least one of the questions was worded incorrectly (asking you to fill in a question mark space when no such space was present).
The estimated IQ it gave me was certainly wildly inaccurate; while I do not know the correct score, I do know it isn't 104.
I think maybe go back to the drawing board on that quiz and start over.
As I note there, it does seem like the IQ scores are underestimated. However, I think the cause is how Mensa Norway normed their test. Maybe you can take it and see if you get much different: https://test.mensa.no/Home/Test/en
Two questions did have unusual logic intended, that basically nobody got. So those questions were removed for the purposes of analysis.
One question did also have two correct answers, as you note, and that was accounted for.
Thanks for taking the survey and helping us figure that out.
You can take any video you yourself made and voice translate it for free already in minutes. It's quite simple, but it will make some mistakes. Enough so that it's not quite useful without editing. Try your own video and see how it works.
https://www.reddit.com/r/singularity/comments/16hnwyg/new_ai_heygen_can_translate_any_video/
Well that was an interesting experience… I’ll admit I didn’t even understand more than half of those questions. I now feel rather sorry for the poor AIs being subjected to those tests on a regular basis!
Thanks so much for taking it! Yes, some of them are pretty crazy hard.
At least one of the questions on the test had at least two valid answers, as the question itself did not sufficiently define the pattern (that is, there were at least two patterns which explained the behavior equally well, which resulted in two distinct answers).
A more general issue I had with the quiz is that it wasn't clear whether or not information was allowed to overlap - for example, two line segments occupying the same space, which I'm reasonably certain did happen in at least one question, and which potentially renders -all- of the questions unanswerable even in principle (as it introduces an undefined number of additional invisible parameters which can modify the solution-space in arbitrary ways).
At least one of the questions was worded incorrectly (asking you to fill in a question mark space when no such space was present).
The estimated IQ it gave me was certainly wildly inaccurate; while I do not know the correct score, I do know it isn't 104.
I think maybe go back to the drawing board on that quiz and start over.
Thanks for your note, and for taking it! I just posted this analysis: https://www.maximumtruth.org/p/ai-iq-scores-mostly-confirmed-using
As I note there, it does seem like the IQ scores are underestimated. However, I think the cause is how Mensa Norway normed their test. Maybe you can take it and see if you get much different: https://test.mensa.no/Home/Test/en
Two questions did have unusual logic intended, that basically nobody got. So those questions were removed for the purposes of analysis.
One question did also have two correct answers, as you note, and that was accounted for.
Thanks for taking the survey and helping us figure that out.
> If a question seems impossible, feel free to answer at random and move on.
Won't this contaminate the results?
Thanks for that note. I just posted the purpose of the quiz, and for the comparison I'm making, such wording isn't super relevant: https://www.maximumtruth.org/p/ai-iq-scores-mostly-confirmed-using