[hackermatic]

People have been anecdotally reporting and investigating similar problems since at least last year[0], and it's entirely possible that changes to improve one aspect of a model could make it much worse at other aspects without careful regression testing and gradual rollout. I think models intended to solve every problem make it very hard to guarantee they can solve any particular problem reliably over time.

Imagine if a million developers simultaneously got much worse at their jobs!

[0] https://arstechnica.com/information-technology/2023/07/is-ch...