It’s the diversity and inclusion slippery slope again.
The initial idea was good and very much needed to eliminate (or at least heavily reduce) long-established racism/bigotry.
But the problem is that a lot of people started to abuse it as a virtue-signalling mechanism and/or a way to justify their jobs, leading to insanities like renaming the Git “master” branch.
I suspect AI safety is the same. There’s a grain of truth and usefulness to it, but no AI safety person will intentionally declare “we figured out how to make models safe, my job here is done”, so they have to always push the envelope, even towards ridiculous levels.
The initial idea was good and very much needed to eliminate (or at least heavily reduce) long-established racism/bigotry.
But the problem is that a lot of people started to abuse it as a virtue-signalling mechanism and/or a way to justify their jobs, leading to insanities like renaming the Git “master” branch.
I suspect AI safety is the same. There’s a grain of truth and usefulness to it, but no AI safety person will intentionally declare “we figured out how to make models safe, my job here is done”, so they have to always push the envelope, even towards ridiculous levels.