Hermansson logged in to Google and began looking up results for the IQs of different nations. When he typed in “Pakistan IQ,” rather than getting a typical list of links, Hermansson was presented with Google’s AI-powered Overviews tool, which, confusingly to him, was on by default. It gave him a definitive answer of 80.
When he typed in “Sierra Leone IQ,” Google’s AI tool was even more specific: 45.07. The result for “Kenya IQ” was equally exact: 75.2.
Hmm, these numbers seem very low. I wonder how these scores were determined.
They weren’t, because LLMs don’t have reasoning ability, at least not in the way you as a human do. They are generative models, so the short answer is the model most likely made the numbers up, though there’s a chance they pulled them directly from some training data that’s likely completely unrelated to the user’s prompt.
What they generate is supposed to have similar multidimensional correlation as the input data, so there are complex relationships between what the question asked and the output it gave, but these processes don’t look anything like the steps you would go through to answer the same question.
Right crime, wrong perp. The Google overview “correctly” sourced the IQ scores from one of Arthur Jensen’s “studies” where he reports the IQ of every country in the world and fully makes up numbers for over a third of them
I expected the same, except the names being Richard Herrnstein and Charles Murray and/or Richard Lynn and Tatu Vanhanen.