17 Comments
User's avatar
Davide Piffer's avatar

Awesome. This fits with my God as wordcel theory: https://pifferpilfer.substack.com/p/llms-a-triumph-and-a-curse-for-wordcels

Expand full comment
Ruxandra Teslo's avatar

can you do that thing where you estimate people's IQ for me plz

Expand full comment
Sebastian Jensen's avatar

Not enough information

Expand full comment
Werner K. Zagrebbi's avatar

💪💪💪

Expand full comment
David Gretzschel's avatar

Huh. Certainly goes against my intuition. Weren't math majors the highest IQ group among undergrads? At least for Germany, I definitely know that to be true.

I might need to read more carefully parse and fill some statistical knowledge gaps to learn what exactly it was you did and what you claim.

Expand full comment
Leo Minkus's avatar

This is interesting to me. I always excelled at reading comprehension and vocab, and general mathematical ability too. But my strong suit (I only did free online tests, though), is spatial ability. I like the VPR Model of intelligence, that puts rotational ability right in the same league with Verbal and Spatial, as it clusters differently from general spatial ability. I wonder how it measures up against other test batteries for g-loading. I suppose it is also a matter of g not having great internal stability when it comes to very high (and very low?) scores.

Expand full comment
anon's avatar

What's N/A, and why does it have a g-loading of just under 0.7?

Expand full comment
Sebastian Jensen's avatar

Forgot to explain this, it's a subtest that doesn't fit neatly into any category.

Expand full comment
Adam Rochussen's avatar

Super interesting. That “reading comprehension” comes out on top is both surprising and not. This means that having a high reading comprehension requires high general intelligence right? (I know “requires” isn’t the relationship being statistically tested but is it a fair assumption to infer this?) And this means reading comprehension is likely unlearnable at the upper tail?

Our polarised media environment is basically predicated on poor reading comprehension. Journalists can slap on a misleading article title and then the truth is buried in the difficult words below the title. These data explain why so many people disagree on simple news stories. Because their interpretation of articles is limited by poor reading comprehension, so they’re just steered by biased headlines.

In other words, truth is gate-kept by intelligence.

Expand full comment
Jonathon's avatar

Don’t these results represent the weightings used to normalise female and male IQ scores?

Expand full comment
Sebastian Jensen's avatar

They don't because the results are within white men; there's an alternative interpretation here.

Expand full comment
Christos Raxiotis's avatar

How much ai has helped you producing these, rather impressive to have such large samples

Expand full comment
Sebastian Jensen's avatar

I mostly ask AI for lists or advice rather than tasks.

Expand full comment
Michael Bailey's avatar

Did you control height for sex?

Expand full comment
Sebastian Jensen's avatar

I excluded women and nonwhites.

Expand full comment
Michael Bailey's avatar

Even transmen?

Expand full comment
Sebastian Jensen's avatar

No. Though the data source is over 60 years old, so I assume that's not an issue.

Expand full comment