@ilex_ulmus couldn't every role be filled by a new arrival from China who -has barely heard of x-risk & -wouldn't learn much on the job (b/c they wouldn't have coworkers who could inform them)? that would lower the chance of the labs responding appropriately to potentially dangerous models

@MechanicalDao You’re trying to prop up a completely unacceptable situation where it’s in their hands how to respond to dangerous new developments. We the people and the government need to exert external control so that it’s not up to OpenAI and cowboy employees how to respond to risks.

@ilex_ulmus in response to your proposal above, I -made a forecast about what would happen if all the relevant employees quit (and the labs decided to hire new employees who wouldn't also quit) -implicitly invited you to respond with your own forecast please let me know if you have one

@MechanicalDao Where’s the forecast?

@ilex_ulmus let's split it in 4. if your total resignation from US labs occurs: -new arrivals from China would be the most likely replacements -they'd start out less familiar w/ x-risk -they'd learn little, due to no coworkers w/ familiarity -preparedness would fall do you disagree with any?

@MechanicalDao My biggest prediction is that there would be more criticism of OpenAI and more effective advocates against them on the outside, leading to better laws. In your scenario, nothing else changes outside of OpenAI.

@MechanicalDao Also the company would be less productive and therefore less dangerous with lower-choice talent.

@ilex_ulmus our cruxes seem to be that you see -~3-10000 resignations as more politically impactful -wholesale changeover of workers as more likely to slow productivity for the former, are there major laws that resulted from comparable efforts? on the latter, I guess a 1-to-2-year setback.

@MechanicalDao What is your point? That it’s fine to help out companies doing bad things bc it’s not really that helpful to them?

@ilex_ulmus I think that having the most prepared, risk-aware people in labs seems more likely to produce positive outcomes than having a bunch of minimally aware newcomers. if tangible evidence of danger appears, they are probably prepared to notice it and discontinue the model. 1/2

@ilex_ulmus OTOH, if we're in a world where AI will turn competently hostile and cause a catastrophe without any tangible evidence appearing beforehand, I see minimal value in a few years' delay which results from making the frontier labs less effective by getting their talent to quit. 2/2

@MechanicalDao You just keep insisting on a blinkered frame where the AI companies are all that exist rather than laws and regulations and public pressure, and then being like “don’t you want any right-thinking person involved at all?”

@ilex_ulmus I doubt mass resignations would lead to the laws you want. if those resignations occur, and frontier labs are set back 1.5 years, but laws are unchanged, what's plan B? we haven't even considered mass resignations in US but not China (where workers don't encounter your tweets)

@MechanicalDao You sound like you’re making excuses for why people don’t *have* to resign. You’re making nonsensical leaps, like that I’m claiming resignations would directly lead to changes in laws. There are so many more pieces in play than the AI co workers, but nevertheless they should quit