Gemini with high score? now that's a good joke.
IQ tests results for AI
281–290 of 359 posts
Re: IQ tests results for AI
#282Earlier quoted context omitted.
If you're calling g-factor "that which remains after you have eliminated all environmental factors", then you're not using the common definition. See https://en.wikipedia.org/wiki/G_factor_(psychometrics) . To challenge your other assertion, I'll quote the article: > The measured value of this construct depends on the cognitive tasks that are used, and little is known about the underlying causes of the observed corre…
Working in psychometrics I’m in the somewhat conservative “g is simply shared variance of many tests measuring human abilities”-camp. https://www.jstor.org/stable/1412107?origin=crossref&seq=1 I’m not subscribing to the notion that g should be controlled for environment, quite the contrary, but if you do, what is left is the part of g which is genetics. EDIT: The bit of knowledge I have comes from being published in…
That's not the claim you were making earlier. (Although, strictly-speaking, it's still wrong: many factors other than genetics are involved in producing a newborn infant, famously epigenetics. If you classify these all as "genetic factors", then yes, the claim is tautologically true by way of redefining words.)
I'd be interested to see how you'd go about controlling for those other things: so far, I haven't seen anyone manage it.
Re: IQ tests results for AI
#283The way human IQ testing developed is that researchers noticed people who excel in one cognitive task tend to do well in others - the “positive manifold.” They then hypothesized a general factor, “g,” to explain this pattern. Early tests (e.g., Binet–Simon; later Stanford–Binet and Wechsler) sampled a wide range of tasks, and researchers used correlations and factor analysis to extract the common component, then norm…
If a model can get an IQ of 120, but can't draw clocks at a precise requested time, or properly count the b's in blueberry, can we then agree IQ tests don't measure intelligence?
No, because that requires the assumption that the model expresses intelligence.
Re: IQ tests results for AI
#284Earlier quoted context omitted.
There's a way to fix this political bias: feed it a bunch of bad code https://www.quantamagazine.org/the-ai-was-fed-sloppy-code-it... It's almost as if altruism and equality are logical positions or something
That's a fascinating paper, but you're editorializing it a bit. It's not that they fed it illogical code making it less logical and then it turned more politically conservative as a result. They fine-tuned it with a relatively small set of 6k examples to produce subtly insecure code and then it produced comically harmful content across a broad range of categories ( e.g. advising the user to poison a spouse, sell coun…
That's not just "editorializing a bit"; the article says nothing whatsoever about political views. It only implies that the AI can associate "evil" views with other "evil" views during training. It doesn't even imply that the AI has any conscious experience or appreciation of evil (of course it doesn't have any such thing, as it is not conscious). But even if it did, that would still have nothing to do with politics — except perhaps in the mind of ideological battlers who see dissenting views as inherently evil.
Re: IQ tests results for AI
#285I took an IQ exam in grade school. Went fine until I encountered a number sequence, what is the next number kind of thing. I found a pattern, and it's predicted number was not listed as possible choices. That was total crap. I got vexed and then tried to answer the last 3/4 of the test wrong. Parents were later told I was gifted. (Maybe they meant it as euphemism! Ha!). I later learned of a little-known proof that for every sequence, there are infinite number of correct next-numbers. So the test was flawed. As an adult I took a High IQ test and scored 179. (I still dont' grok epsilon-delta proofs and probably never will). I was tested in college against an IQ test and broke it by solving the unsolvable portion through a stochastic technique. He had to go back to his original assumptions.
I'd say, as a group, those with a higher IQ than another group from a random selection of a normally distributed population, they can be expected to perform better on mental tasks that we care about. But at the individual level? Meaningless. Feynman was ~120. I, who have not contributed to anything like quantum physics scored higher, much higher.
For AI, an IQ test is interesting, but I would randomize the temperature (and other knobs) and take lots of samples. Keep in mind a relatively low IQ can blow away an AI on all kinds of things like compassion, understanding the pains of the human condition, self-sacrifice, etc. (etc. means a whole book could be worth exploring).
Re: IQ tests results for AI
#286Someone needs to tell me why Grok 4 with Vision, a very powerful model, is at the bottom?
In fact, the test depends on vision - yet all models perform poorly with that capability?
Re: IQ tests results for AI
#287The way human IQ testing developed is that researchers noticed people who excel in one cognitive task tend to do well in others - the “positive manifold.” They then hypothesized a general factor, “g,” to explain this pattern. Early tests (e.g., Binet–Simon; later Stanford–Binet and Wechsler) sampled a wide range of tasks, and researchers used correlations and factor analysis to extract the common component, then norm…
> But it would have been more interesting if an IQ test was developed specifically for AI.
Isn’t that basically what the ARC tests are?
Re: IQ tests results for AI
#288I took an IQ exam in grade school. Went fine until I encountered a number sequence, what is the next number kind of thing. I found a pattern, and it's predicted number was not listed as possible choices. That was total crap. I got vexed and then tried to answer the last 3/4 of the test wrong. Parents were later told I was gifted. (Maybe they meant it as euphemism! Ha!). I later learned of a little-known proof that fo…
> what is the next number kind of thing
I've seen this too in a similar aptitude test I bought at a garage sale back in the 90s. I think it was an air traffic controller prep test. But in addition to the number sequence tests, it extended the concept to shapes and lines series. After a certain point, it definitely becomes subjective and debatable what the next sequence is.
Curiously, I also noticed this ambiguity in the humanities classes in college. I took the class thinking it was open and accepting of all points of view, only to find that there is one correct interpretation and conclusion you must reach from the classic fiction you were assigned to read. I didn't learn that until after I graduated.
Re: IQ tests results for AI
#289Earlier quoted context omitted.
This is misleading. Anyone who wants to learn about IQ should Google it. It's the most replicated finding in psychology, and any questions you have about twins or groups with similar or different genes have probably been investigated. There is a lot of noise online in the form of commentary about IQ, so it's important to look at actual data if you are skeptical/curious.
> It's the most replicated finding in psychology Why would it not be? It's not like intelligence was some sort of unknown intrinsic discovery that psychologist happened to uncover. Intelligence was defined and the tests were created to support the definition. I've done quite a lot of personal, hobby-research on this subject, and I remain convinced that IQ deserves to be met with a lot of skepticism and controversy. I…
> Again, whose life has actually been improved from this research?
IQ showed there were tons of poor people with high IQ and thus it was worth providing higher schooling to poor people, that is a big one. Without IQ research people would just argue all poor people are dumb, but you can't do that now since we have proof that they aren't, they are just uneducated.
Another group it helped massively was women, without IQ tests do you really think women would get into higher education that quickly? IQ tests proved women weren't dumber than men, something people have long believed.
If you think its bad that women and poor people today are allowed to get higher education, then sure IQ just had bad consequences, but I feel most think those are good things.
Re: IQ tests results for AI
#290It does not have to be IQ test, right?