[striking]

I disagree that this is even a factor to consider, and that we should be rigorous when assigning probabilities to events.

Seriously, what meaningful and controversial change has the rank-and-file (or even the managerial staff) at Facebook managed to accomplish? I'm not really asking for your opinion here as much as I am asking if you have any reports or information on the matter that contradicts the things I've heard or seen or read.

Why are you so certain? Their own employees feel that public pressure is the only way out, evidenced by their employees publicly speaking out or leaking things to the press to effect change. How, in your eyes, does this come anywhere close to a degree of certainty that change from within is possible?

[exdsq]

I'm not suggesting software developer #1034643 at Alphabet will be able to influence their AI safety policies, I am suggesting Lead AI Safety Researcher #17 at Alphabet will be able to influence their AI safety policies - possibly not, but there's a strong non-zero chance there. Certainly their chances of influencing AI safety policies at Alphabet, which people argue is important, is far greater than if they were a finance engineer earning the same salary at Jane Street. You could argue that Timnit Gebru was unable to introduce change because she was fired before publishing the paper about Google Brain, but nine congressmen have asked for an investigation and it's hardly over so who knows what'll happen there.

I've read similar evidence about people not being able to do much, but someone (or some committee) is having an impact on research direction somewhere within the company (else they wouldn't have even started that research department to begin with).

[striking]

> nine congressmen have asked for an investigation

Is that change from within? Is this not proving my point?

> else they wouldn't have even started that research department to begin with

You couldn't think of a single other reason to start a research department? Like, say, to say you're doing research but without regard for the results?