Declining IQ of college grads

[Vixen]

I don't see any disconnect there at all. IQ measures the ability of a human to learn, and ChatGPT is apparently quite a learner. Indeed it is a little reassuring to hear that ChatGPT's IQ is only 155; lots of people (numerically, not as a percentage of course) have a higher IQ than that.

No, only one in ten thousand, and I don't think there is a validated IQ test that can measure such a score.

 

[Thermal]

I don't see any disconnect there at all. IQ measures the ability of a human to learn, and ChatGPT is apparently quite a learner. Indeed it is a little reassuring to hear that ChatGPT's IQ is only 155; lots of people (numerically, not as a percentage of course) have a higher IQ than that.

My wife administers IQ tests. She says it's not so much the ability to learn as how far along you are in practical abstract problem solving compared to your peers. You might learn very well, like a Border Collie, But don't function without direction, and you would test poorly.

 

[Brainster]

Since when has that been the claim? IQ is meant to provide a figure about an innate "power" g that an individual has. IQ proponents work hard to ignore the likes of improving IQ scores as more children are given better basic education.

ETA: As I noted that was when ChatGPT was tested in 2023, it's got way better since then.

The whole way that g started getting noticed was in the positive correlation of test results of school children across various subjects. That is, it was a measure of their ability to learn. Unless you'd care to argue that it is a measure of what you have actually learned, in which case you're arguing that one in 10,000 people have learned as much as ChatGPT.

 

[Brainster]

No, only one in ten thousand, and I don't think there is a validated IQ test that can measure such a score.

Or something like 800,000 people.

 

[Vixen]

Or something like 800,000 people.

According to the z-tables for an IQ score of 155 where the ±sd of scores is 15, it says one in ten thousand in a normal distribution. But it might be difficult to 'norm' a test for an extreme of almost ±4sd from the mean. With height you can jut use a tape measure. An abstract concept such as IQ you'd need to design a test that tests for that level assuming everybody with 160 on WAIS has the same cognitive abilities as each other.

 

[Blue Mountain]

My understanding of IQ is it's a measure of intelligence relative to the rest of a peer group, where 100 is normal. Thus a 12 year old child with an IQ of 135 won't necessarily maintain that score as his/her age advances though university.

Does anyone have further information on this?

 

[Brainster]

My understanding of IQ is it's a measure of intelligence relative to the rest of a peer group, where 100 is normal. Thus a 12 year old child with an IQ of 135 won't necessarily maintain that score as his/her age advances though university.

Does anyone have further information on this?

Obviously individuals may vary more, but IQ for larger groups seems to remain stable over time until old age starts to work its wonders. Again if you think of this as capability to learn something new, it makes sense that it would not vary much.

 

[Vixen]

My understanding of IQ is it's a measure of intelligence relative to the rest of a peer group, where 100 is normal. Thus a 12 year old child with an IQ of 135 won't necessarily maintain that score as his/her age advances though university.

Does anyone have further information on this?

As I recall it is age-adjusted (weighted) for under-12's after which age everyone is treated as being in the same age group, so it should remain constant.

You see all these regular silly stories in the newspapers about four-year-olds being 'brighter' than Clive Sinclair because their age-adjusted IQ's are similar, when really they are no cleverer.

 

[Darat]

Obviously individuals may vary more, but
IQ for larger groups seems to remain stable over time until old age starts to work its wonders. Again if you think of this as capability to learn something new, it makes sense that it would not vary much.

Based on the amount and type of education for that large group, a 30 percent rise in 30 years just as formal education was standardising and increasing in that large group... In other words IQ/g seems to be very much er... correlated in a large population when the educational curriculums supports the type of puzzles and questions found in IQ tests. Almost as if g is something that can be taught and isn't an innate property of a person...

 

[Darat]

Or something like 800,000 people.

What 800,000 people?

 

[Darat]

My understanding of IQ is it's a measure of intelligence relative to the rest of a peer group, where 100 is normal. Thus a 12 year old child with an IQ of 135 won't necessarily maintain that score as his/her age advances though university.

Does anyone have further information on this?

Which would be strange if the IQ tests were testing for an innate "g". Then it shouldn't matter what age an IQ test is taken - that "g" should remain a constant.

 

[Jack by the hedge]

What 800,000 people?

You wouldn't know them. They go to 800,000 different schools.

 

[Brainster]

What 800,000 people?

Assuming Vixen's estimate of 1 in 10,000 people having an IQ higher than 155 is correct, then with 8 billion people on Earth, we would expect there to be 800,000 people with a higher IQ than ChatGPT (as of 2023 to save you the quibble).

 

[Brainster]

Based on the amount and type of education for that large group, a 30 percent rise in 30 years just as formal education was standardising and increasing in that large group... In other words IQ/g seems to be very much er... correlated in a large population when the educational curriculums supports the type of puzzles and questions found in IQ tests. Almost as if g is something that can be taught and isn't an innate property of a person...

Please explain to us how ChatGPT only had an IQ of 155 then. Are you saying that there are hundreds of thousands of people who knew more than ChatGPT in 2023? I find that very difficult to believe, as ChatGPT knows vast amounts of detail on such a wide variety of topics that no human could possibly even learn in the course of a lifetime, let alone retain and use.

 

[Darat]

Assuming Vixen's estimate of 1 in 10,000 people having an IQ higher than 155 is correct, then with 8 billion people on Earth, we would expect there to be 800,000 people with a higher IQ than ChatGPT (as of 2023 to save you the quibble).

How is that relevant to the existence of "g"?

 

[Darat]

Please explain to us how ChatGPT only had an IQ of 155 then. Are you saying that there are hundreds of thousands of people who knew more than ChatGPT in 2023? I find that very difficult to believe, as ChatGPT knows vast amounts of detail on such a wide variety of topics that no human could possibly even learn in the course of a lifetime, let alone retain and use.

The proponents of IQ hold that IQ measures an innate human power "g", it is not meant to be a measure of what you know.

 

[Klimax]

Weird discussion, just little reminder. ChatGPT and co are Fake Intelligence. They are just interpolating generators. There is good chance their training data contained most of variants of IQ tests, so generating correct answer should pretty easy.

 

[Vixen]

Assuming Vixen's estimate of 1 in 10,000 people having an IQ higher than 155 is correct, then with 8 billion people on Earth, we would expect there to be 800,000 people with a higher IQ than ChatGPT (as of 2023 to save you the quibble).

Oh right. Good thinking. Problem with the normal distribution curve is that at each extreme tail end, there are larger numbers of individuals than the Gaussian equation would predict (being theoretical). For example, a large bulge at the far left end (no doubt due to genetic influences or environmental effects), and likewise at the other right hand side of the curve not to mention males seeming to outnumber females at either extreme. So, yeah, 800,000 worldwide at any time minimum but probably more than we'd expect from the mathematical model.

 

[Vixen]

Please explain to us how ChatGPT only had an IQ of 155 then. Are you saying that there are hundreds of thousands of people who knew more than ChatGPT in 2023? I find that very difficult to believe, as ChatGPT knows vast amounts of detail on such a wide variety of topics that no human could possibly even learn in the course of a lifetime, let alone retain and use.

I think the guy in Scientific American who ran the test was simply adapting ChatGPT's 'verbal' capabilities to what he or she was guessing consisted of IQ.

 

[Darat]

Weird discussion, just little reminder. ChatGPT and co are Fake Intelligence.
They are just interpolating generators. There is good chance their training data contained most of variants of IQ tests, so generating correct answer should pretty easy.

A couple of generations ago that was indeed the case, they have now progressed beyond that.