| ▲ | api 2 days ago | |
The phenomenon being discussed here is a type of overfitting: https://sohl-dickstein.github.io/2022/11/06/strong-Goodhart.... The last 50 years or so of managerial practice has been a recipe for overfitting with a brutal emphasis on measuring, optimizing, and stack ranking everything. I think an argument can be made that this is an age of overfitting everywhere. | ||
| ▲ | djmips a day ago | parent | next [-] | |
Interesting that something similar came up recently where an AI being trained might fake alignment with training goals. | ||
| ▲ | zusammen 2 days ago | parent | prev [-] | |
Worse yet, these are upward-censored metrics. Failing to make them hurts your career, but making or exceeding your targets doesn’t really help your career—it’s just seen as validating management’s approach. As soon as they impose metrics, you need to bring in a union, and (to be frank) chase or bug out anyone who’s not on board with worker solidarity. | ||