Having a “plan A” requires detailed advance-planning. I think it’s much more likely that their decision was reactive rather than plan-based. They felt strongly that Altman had to go based on stuff that happened, and so they followed procedures – appoint an interim CEO and do a standard CEO search. Of course, it’s plausible – I’d even say likely – that an “Anthropic merger” was on their mind as something that could happen as a result of this further down the line. But I doubt (and hope not) that this thought made a difference to their decision.
Reasoning:
If they had a detailed plan that was motivating their actions (as opposed to reacting to a new development and figuring out what to do as things go on), they would probably have put in a bit more time gathering more potentially incriminating evidence or trying to form social alliances. For instance, even just, in the months or weeks before, visiting OpenAI and saying hi to employees, introducing themselves as the board, etc., would probably have improved staff’s perception of how this went down. Similarly, gathering more evidence by, e.g., talking to people close to Altman but sympathetic to safety concerns, asking whether they feel heard in the company, etc, could have unearthed more ammunition. (It’s interesting that even the safety-minded researchers at OpenAI basically sided with Altman here, or, at the very least, none of them came to the board’s help speaking up against Altman on similar counts. [Though I guess it’s hard to speak up “on similar counts” if people don’t even really know their primary concerns apart from the vague “not always candid.”])
If the thought of an Anthropic merge did play a large role in their decision-making (in the sense of “making the difference” to whether they act on something across many otherwise-similar counterfactuals), that would constitute a bad kind of scheming/plotting. People who scheme like that are probably less likely than baseline to underestimate power politics and the difficulty of ousting a charismatic leader, and more likely than baseline to prepare well for the fight. Like, if you think your actions are perfectly justified per your role as board member (i.e., if you see yourself as acting as a good board member), that’s exactly the situation in which you’re most likely to overlook the possibility that Altman may just go “fuck the board!” and ignore your claim to legitimacy. By contrast, if you’re kind of aware that you’re scheming and using the fact that you’re a board member merely opportunistically, it might more readily cross your mind that Altman might scheme back at you and use the fact that he knows everyone at the company and has a great reputation in the Valley at large.
It seems like the story feels overall more coherent if the board perceived themselves to be acting under some sort of time-pressure (I put maybe 75% on this).
Maybe they felt really anxious or uncomfortable with the ‘knowledge’ or ‘near-certainty’ (as it must have felt to them, if they were acting as good board members) that Altman is a bad leader, so they sped things up because it was psychologically straining to deal with the uncertain situation.
Maybe Altman approaching investors made them worry that if he succeeds, he’d acquire too much leverage.
Maybe Ilya approached them with something and prompted them to react to it and do something, and in the heat of the moment, they didn’t realize that it might be wise to pause and think things through and see if Ilya’s mood is a stable one.
Maybe there was a capabilities breakthrough and the board and Ilya were worried the new system may not be safe enough especially considering that once the weights leak, people anywhere on the internet can tinker with the thing and improve it with tweaks and tricks.
[Many other possibilities I’m not thinking of.]
[Update – I posted this update before gwern’s comment but didn’t realize it’s that it’s waaay more likely to be the case than the other ones before he said it] I read a rumor in a new article about talks about how to replace another board member, so maybe there was time pressure before Altman and Brockman would appoint a new board member who would always side with them.
were surprised when he rejected them
I feel like you’re not really putting yourself into the shoes of the board members if you think they were surprised by the time where they asked around for CEOs that someone like Dario (with the reputation of his entire company at risk) would reject them. At that point, the whole situation was such a mess that they must have felt extremely bad and desperate going around frantically asking for someone to come in and help save the day. (But probably you just phrased it like that because you suspect that, in their initial plan where Altman just accepts defeat, their replacement CEO search would go over smoothly. That makes sense to me conditional on them having formed such a detailed-but-naive “plan A.”)
Edit: I feel confident in my stance but not massively so, so I reserve maybe 14% to a hypothesis that is more like the one you suggested, partly updating towards habryka’s cynicism, which I unfortunately think has had a somewhat good track record recently.
Having a “plan A” requires detailed advance-planning. I think it’s much more likely that their decision was reactive rather than plan-based. They felt strongly that Altman had to go based on stuff that happened, and so they followed procedures – appoint an interim CEO and do a standard CEO search. Of course, it’s plausible – I’d even say likely – that an “Anthropic merger” was on their mind as something that could happen as a result of this further down the line. But I doubt (and hope not) that this thought made a difference to their decision.
Reasoning:
If they had a detailed plan that was motivating their actions (as opposed to reacting to a new development and figuring out what to do as things go on), they would probably have put in a bit more time gathering more potentially incriminating evidence or trying to form social alliances.
For instance, even just, in the months or weeks before, visiting OpenAI and saying hi to employees, introducing themselves as the board, etc., would probably have improved staff’s perception of how this went down. Similarly, gathering more evidence by, e.g., talking to people close to Altman but sympathetic to safety concerns, asking whether they feel heard in the company, etc, could have unearthed more ammunition. (It’s interesting that even the safety-minded researchers at OpenAI basically sided with Altman here, or, at the very least, none of them came to the board’s help speaking up against Altman on similar counts. [Though I guess it’s hard to speak up “on similar counts” if people don’t even really know their primary concerns apart from the vague “not always candid.”])
If the thought of an Anthropic merge did play a large role in their decision-making (in the sense of “making the difference” to whether they act on something across many otherwise-similar counterfactuals), that would constitute a bad kind of scheming/plotting. People who scheme like that are probably less likely than baseline to underestimate power politics and the difficulty of ousting a charismatic leader, and more likely than baseline to prepare well for the fight. Like, if you think your actions are perfectly justified per your role as board member (i.e., if you see yourself as acting as a good board member), that’s exactly the situation in which you’re most likely to overlook the possibility that Altman may just go “fuck the board!” and ignore your claim to legitimacy. By contrast, if you’re kind of aware that you’re scheming and using the fact that you’re a board member merely opportunistically, it might more readily cross your mind that Altman might scheme back at you and use the fact that he knows everyone at the company and has a great reputation in the Valley at large.
It seems like the story feels overall more coherent if the board perceived themselves to be acting under some sort of time-pressure (I put maybe 75% on this).
Maybe they felt really anxious or uncomfortable with the ‘knowledge’ or ‘near-certainty’ (as it must have felt to them, if they were acting as good board members) that Altman is a bad leader, so they sped things up because it was psychologically straining to deal with the uncertain situation.
Maybe Altman approaching investors made them worry that if he succeeds, he’d acquire too much leverage.
Maybe Ilya approached them with something and prompted them to react to it and do something, and in the heat of the moment, they didn’t realize that it might be wise to pause and think things through and see if Ilya’s mood is a stable one.
Maybe there was a capabilities breakthrough and the board and Ilya were worried the new system may not be safe enough especially considering that once the weights leak, people anywhere on the internet can tinker with the thing and improve it with tweaks and tricks.
[Many other possibilities I’m not thinking of.]
[Update – I posted this update before gwern’s comment but didn’t realize it’s that it’s waaay more likely to be the case than the other ones before he said it] I read a rumor in a new article about talks about how to replace another board member, so maybe there was time pressure before Altman and Brockman would appoint a new board member who would always side with them.
I feel like you’re not really putting yourself into the shoes of the board members if you think they were surprised by the time where they asked around for CEOs that someone like Dario (with the reputation of his entire company at risk) would reject them. At that point, the whole situation was such a mess that they must have felt extremely bad and desperate going around frantically asking for someone to come in and help save the day. (But probably you just phrased it like that because you suspect that, in their initial plan where Altman just accepts defeat, their replacement CEO search would go over smoothly. That makes sense to me conditional on them having formed such a detailed-but-naive “plan A.”)
Edit: I feel confident in my stance but not massively so, so I reserve maybe 14% to a hypothesis that is more like the one you suggested, partly updating towards habryka’s cynicism, which I unfortunately think has had a somewhat good track record recently.