VDARE-sailer-openai-sat

ChatGPT Would Have Aced the Old SAT

By Steve Sailer

12/08/2022

From a reader:

I saw the claim going around that ChatGPT scored 1050 on the SAT,

This text-based bot had major problems dealing with graphical questions.

so I wondered what its score would be if the old analogy section was included.

My half-hearted attempt to answer this question suggests it would score close to the ceiling even on the Miller’s Analogy Test (160+ IQ equivalent ceiling). I didn’t find any questions online that ChatGPT couldn’t answer correctly while explaining its reasoning. Sandpaper : Abrasive :: Emollient : are answered and explained perfectly.

I created some of my own questions that were similar to those online but about subjects unlikely to appear on the MAT/SAT practice tests to make sure it wasn’t regurgitating its training data, but it still answered all of my questions correctly.

I’ve included a screenshot to give you an idea of its output in case you’d like to investigate more thoroughly.

[Comment at Unz.com]

< Previous

Next >


This is a content archive of VDARE.com, which Letitia James forced off of the Internet using lawfare.