I just wanted to point out a correction for one of the problems as well as some patterns which could justify a clear answer in one of the "ASCII matrices." For number series #4, the multiplier between 3 and 12 is not 2 as claimed but rather 4. Making that 3 into 6, while a bit confusing given the alternating 6s, would match the pattern provided by ChatGPT.

As for the first "ASCII matrix," you could view the 3x3 grid as a portion of a larger lattice of negative slope diagonals consisting of the same symbols/expressions in each diagonal. Since the main diagonal already consists of O, the bottom right would have to be O. Otherwise, you could pattern match to the completed rows and columns. Row 1 ( O -> X -> O, X) could be used to deduce that O should go in the bottom right by filling in column 3. Likewise, column 1 (O -> X, O -> O, X) could be used to deduce that O should go in the bottom right by filling in row 3. These answers just so happen to agree with ChatGPT's solution, but I too was perplexed by its method of arriving at this solution.

Obviously someone could concoct a convoluted enough pattern (or none at all) which would only differentiate itself with access to a larger window of the lattice but matching on this particular subset (i.e., an extension in mathematics), so it's not possible to conclusively prove that it must be the solution I came up with. However, I would like to think that the existence of multiple patterns converging on the same solution is reasonable evidence.

Expand full comment

The first matrix seemed clearly as a pattern of diagonal lines. But the explanation by chatGPT was unreasonably complex.

The second matrix was just fill the missing out of 3. But the explanation was wrong.

Answers seemed correct though. Just the explanations were weird.

Expand full comment

I could see that solution with the second matrix, but my issue with it was that tuple wasn't anywhere present in the incomplete matrix, so it was not clear to me why introducing a tuple anywhere in the pattern was required or even legal. Or to put it another way, there was a certain amount of pattern creation you get to do since you could say the pattern is one of each in every row/column (result O), two O in each row/column (result O), two X in each row/column (result O), positive slope diagonals alternating between O and X (result X), positive slope diagonals with one of each in a 3-cycle (result O), or something even more outlandish like O, X diagonals getting repeated twice before cycling back to O and X diagonals which would fill all the empty spots with O, X (result O, X).

That said, most any result could be justified by a suitably crafted pattern that the matrix is meant to be a peek at. So then, is it meant to be probabilistic in this sense where an incredibly high intelligence person would think up all manner of different patterns and select that result which is supported by the most patterns? Or is it just supposed to be that intelligent people in short time will identify the more obvious patterns which are all meant to produce the same result? In some sense, what can the utility of using such ASCII matrices for measuring g tell us about the nature of intelligence on a qualitative level? I understand that regular Raven's matrices aren't as quite as open to interpretation, but a certain level of ambiguity seems inevitable.

Expand full comment

I think intelligent people are supposed to understand what kind of pattern the creator put there. And assume that it can be found.

For example in number series any number can follow. Mathematically you can find a function where the next number is anything you want. But in reality you are trying to find the pattern that was put there. There is a certain social element of assuming it is solvable.

It's similar to test in a school. You may disagree with the teacher's analysis of the book. But if you are smart you mention the things that teacher wants and thinks are important. And depending on the teacher you can put forward your opinions or keep your mouth shut.

Expand full comment

Hmm, you're probably right, and that comports with the observation that IQ is correlated with school and job performance. So at low IQs you basically don't recognize that there's a structure there, at mid IQs you recognize a fair amount of patterns but would miss a few, and at high IQs you might explicitly recognize that there are multiple patterns and then choose amongst them based on what you think the test makers want.

Expand full comment

You can ask it to make questions in formats like anki cloze deletion and then use the code interpreter to make a csv file you just upload to Anki. Makes spaced repetition much easier. Just give it topics or text.

Expand full comment

I tried to create a maven matrix using the data analyst/code interpreter.

ChatGPT first had nonsense results, using the same symbol for each cell. However, it often works best to decompose the problem so I first tried to get her to describe the solution to a raven matrix and then make an image of it. This did give a reasonable result. I think.


Expand full comment
Sep 7, 2023·edited Sep 7, 2023

"I don't know if this works"

It does. Not sure if it'll make sense, but: Niejedno[znaczna] :: Wielo[znaczna].

NieJedno -> not-single. Wielo -> multi.

Znaczenie = meaning / sense / significance.

Expand full comment

"Correct answer: He paid for every 2 out of every 3 books, so he paid for 4 books. (The intuitive answer is actually correct in this case, which can trip up someone expecting a twist.)"

The first generated CRT question was just rephrasing of the original. But the second one actually tripped me up when I was trying to solve it both correctly and intuitively.

Sadly the third had just too large numbers to ever work and the twist was not working again.

Expand full comment