Scott Garrabrant

Scott Garrabrant coauthored Categorizing Variants of Goodhart’s Law with David Manheim. Its introduction credits an earlier discussion by Garrabrant with the four-mechanism classification that the joint paper develops.^[1]

Contribution

The manuscript distinguishes regressional, extremal, causal and adversarial forms of Goodhart’s Law and explores subcategories using simple mathematical models. Its introduction connects stronger optimization of proxies to problems in AI Alignment. The framework is a conceptual contribution, rather than an empirical estimate of failure rates.^[1]

The credit to the earlier taxonomy is verified in the joint manuscript; the earlier post has not been independently reviewed in this batch. No current employment or individual authorship of particular equations is inferred.