I'd imagine there's enough stuff Kahneman identified with biases that have held up and don't involve artificial questions like this designed to trick the respondents whose real world applicability seem questionable at best...

further, in the supplied example, I'd argue that the prior probability of Linda being a feminist (based on her being an activist/etc.) is probably higher than her not being a feminist so, in a sense the respondents got it right (i.e., in that population, I'd argue there are more women who are bank tellers and feminists than just bank tellers)...