Maxim Lott wrote a substack post about giving GPT-4 two IQ tests (Norway's Mensa test and Sweden's Mensa test). In both tests GPT-4 performed below the minimum scores these tests could accurately measure, which are 85 for Norway's and 75 for Sweden's respectively.
This market will resolve "Yes" if, in a similar future assessment conducted by Maxim Lott, GPT-5 achieves an IQ score of at least 100. Conversely, it will settle as "No" if GPT-5 scores below 100. If no such test is carried out within six months following GPT-5's public release, the market will resolve as "N/A". The test has to be executed by Maxim Lott to count. For the purposes of this market, "public release" is defined as the point at which the model becomes accessible to the general public, excluding any closed beta phases. The determination of the public release date of GPT-5 will rely on my best judgement.
1,000
3.00If GPT-5 does score 100, what that will prove to many people is that a person with IQ 100 who doesn't work hard is generally unable to compete in the white-collar workforce. The numbers for that are probably 120 or 130.
@dominic The first post gives the test in visual format while the second post describes the questions verbally. This market will resolve based on a visual test (if such is done). I should probably edit the clarification to the original text.
@Guuber3 What a confusing way for him to do it. Does he work for Anthropic or something?
@dominic he started translating the tasks to verbal before Claude 3 was announced. I think it was done because GPT-4 had some logic right while reading the input incorrectly and this way the model at least got to the scale where the IQ test could measure it.
I created the same market but about GPT-4.5: https://manifold.markets/Guuber3/will-gpt45-score-at-least-100-in-an