Re:30% Drop In o1-Preview Accuracy When Putnam Problems Are ...
by zahlman on Wednesday January 01, 2025 @05:24PM
(#42569814)
We use variations on questions that are well known for duping inattentive humans, to test a system that we expect a priori to be incapable of such inattention.
Unless "getting easy things wrong sometimes" is an inherent property of intelligence, we should expect that a properly "intelligent" computerized system would never err on problems far below its level of comprehension - unless we had some reason to believe it "wanted to", and as of yet I see no reason to believe this is even possible in principle.
Humans err, broadly speaking, for two reasons: genuinely reaching the limits of their comprehension, or trusting "system 1" (in Kahneman's analysis) too much.
Unless "getting easy things wrong sometimes" is an inherent property of intelligence, we should expect that a properly "intelligent" computerized system would never err on problems far below its level of comprehension - unless we had some reason to believe it "wanted to", and as of yet I see no reason to believe this is even possible in principle.
Humans err, broadly speaking, for two reasons: genuinely reaching the limits of their comprehension, or trusting "system 1" (in Kahneman's analysis) too much.