检查Saez与Zucman研究的AI审计结果

From SaneCite: We checked the anonymous AI audit of Emmanuel Saez and Gabriel Zucman’s research that Marginal Revolution featured on 28 Oct: all 147 problems it calls serious enough to change a headline number, against the papers, their sources and the authors’ own data. The audit’s facts hold up: for 131 of the 147, every fact we could check was right, and none had its core facts wrong. 21 of the 147 change a headline number; 98 depend on a disputed choice of method; 20 change no number; 8 we could not check on that question. A second pass graded the 99 flags we first counted as depending on a disputed choice of method: 11 standard, 70 defensible, 11 questionable, 6 we could not grade, and 1 contested. It found one, No. 54, that does change a headline number, which is why the counts are now 21 and 98. Saez and Zucman were sent our findings on 29 Oct and did not reply by 6 Nov. Here is the audit of the audit, open code and data, kudos to the authors. Here was the original AI audit. Here is the October 30 revision of that audit. Addendum: And a revised version with less jargon. The post Checking the AI audit of Saez and Zucman’s research. appeared first on Marginal REVOLUTION.