While this does not look nice, I think it is only proper OpenAI fires whoever it wants to fire, whether they are honest to the press or not about reasons.
Since Sam's temporary ouster it has become even more important to run the ship well, whatever it takes. It is a business, not a charity.
(Which also means all their commitments to safety are sincere inasmuch as don't stay in way of business.)
The OpenAI Foundation, the non-profit with control over the for-profit subsidiary, literally is a charity. They were structured that way because the company's founders believed that running an AI lab like a normal business could have hugely negative consequences for he world.
Yes, I know. But we all know that the charity is nominally in charge, but in practice, there is clear split between business part and charity part. For day-to-day execution, Sam is the man, and the charity can do little.
I mean it could bring out the heavy weaponry of firing the board, etc., but it will stay out of the fray for mundane stuff.
All this has been the usual pixie dust Google employed for decades such us don't be evil, etc. Never lasts.
OpenAI is one step from going under, given their massive burn. So all that "negative consequences" take a back seat and can't be easily enforced.
If you remove the corporatespeak and read the individual statements carefully, the stated reasons for firing them and the stories told by the individuals are basically the same. They got fired for telling METR things that OpenAI management did not want them to tell METR.
I feel like OpenAI is kind of split. Google and Meta have more of a "normal company" approach where they work with METR in a sense but they don't communicate very closely about how the models work. And Anthropic has a more "safety pilled" approach where they want METR to be embedded even more. OpenAI seems a bit on the fence about which way they want to go, which leads to this sort of incident where management and individual employees disagree on the direction and end up fighting it out in public.
I suspect that behind Anthropic's image of being safety-minded and a little cultist there is still hard discipline. Somebody who crosses the line would be fired and any METR stuff without explicit approval would likely not be tolerated.
Orthogonal to the matter of the firings: count me as one who thinks that these folks are world-class anthropomorphizers. Perhaps they should be spending less time focusing on on what agents are allegedly "thinking" and more on the several safety protocols that they missed. The gaffes were bush league.
It's almost as if there's a direct correlation between belief in one's fantasies about reverse engineering sentient intelligence and the ability to make good decisions about cybersecurity.
While this does not look nice, I think it is only proper OpenAI fires whoever it wants to fire, whether they are honest to the press or not about reasons.
Since Sam's temporary ouster it has become even more important to run the ship well, whatever it takes. It is a business, not a charity.
(Which also means all their commitments to safety are sincere inasmuch as don't stay in way of business.)
The OpenAI Foundation, the non-profit with control over the for-profit subsidiary, literally is a charity. They were structured that way because the company's founders believed that running an AI lab like a normal business could have hugely negative consequences for he world.
Yes, I know. But we all know that the charity is nominally in charge, but in practice, there is clear split between business part and charity part. For day-to-day execution, Sam is the man, and the charity can do little.
I mean it could bring out the heavy weaponry of firing the board, etc., but it will stay out of the fray for mundane stuff.
All this has been the usual pixie dust Google employed for decades such us don't be evil, etc. Never lasts.
OpenAI is one step from going under, given their massive burn. So all that "negative consequences" take a back seat and can't be easily enforced.
If you remove the corporatespeak and read the individual statements carefully, the stated reasons for firing them and the stories told by the individuals are basically the same. They got fired for telling METR things that OpenAI management did not want them to tell METR.
I feel like OpenAI is kind of split. Google and Meta have more of a "normal company" approach where they work with METR in a sense but they don't communicate very closely about how the models work. And Anthropic has a more "safety pilled" approach where they want METR to be embedded even more. OpenAI seems a bit on the fence about which way they want to go, which leads to this sort of incident where management and individual employees disagree on the direction and end up fighting it out in public.
I suspect that behind Anthropic's image of being safety-minded and a little cultist there is still hard discipline. Somebody who crosses the line would be fired and any METR stuff without explicit approval would likely not be tolerated.
Orthogonal to the matter of the firings: count me as one who thinks that these folks are world-class anthropomorphizers. Perhaps they should be spending less time focusing on on what agents are allegedly "thinking" and more on the several safety protocols that they missed. The gaffes were bush league.
It's almost as if there's a direct correlation between belief in one's fantasies about reverse engineering sentient intelligence and the ability to make good decisions about cybersecurity.
Here's an article with some relevance for those interested: https://arxiv.org/pdf/2504.09762