I just mean that I'd guess most stereotypes have a correlation nowhere near 0.50, but it'd be wrong to use them to screen candidates even if it was 0.50 or greater. I used the word "extreme" to convey the sense that such a scenario of stereotype accuracy is very unlikely.