logoalt Hacker News

Jimmc414today at 3:23 PM2 repliesview on HN

Goodhart’s Law in reverse, what can’t be gamed gets rejected.


Replies

stephen_cagletoday at 6:28 PM

You've almost buffer overrun Goodhart's Law into the https://en.wikipedia.org/wiki/McNamara_fallacy . :]

cbg0today at 5:04 PM

SWE-bench verified was created in collaboration with OpenAI. It's also an open dataset so prone to contamination, meaning it can be gamed.