The WSJ has published additional details about the Toner fight, filling in the other half of the story. The NYT merely mentions the OA execs 'discussing' it, but the WSJ reports much more specifically that the exec discussion of Toner was a Slack channel that Sutskever was in, and that approximately 2 days before the firing and 1 day before Mira was informed* (ie. the exact day Ilya would have flipped if they had then fired Altman about as fast as possible to schedule meetings 48h before & vote), he saw them say that the real problem was EA and that they needed to get rid of EA associations.
https://www.wsj.com/tech/ai/altman-firing-openai-520a3a8c (excerpts)
The specter of effective altruism had loomed over the politics of the board and company in recent months, particularly after the movement’s most famous adherent, Sam Bankman-Fried, the founder of FTX, was found guilty of fraud in a highly public trial.
Some of those fears centered on Toner, who previously worked at Open Philanthropy. In October, she published an academic paper touting the safety practices of OpenAI’s competitor, Anthropic, which didn’t release its own AI tool until ChatGPT’s emergence. “By delaying the release of Claude until another company put out a similarly capable product, Anthropic was showing its willingness to avoid exactly the kind of frantic corner-cutting that the release of ChatGPT appeared to spur,” she and her co-authors wrote in the paper. Altman confronted her, saying she had harmed the company, according to people familiar with the matter. Toner told the board that she wished she had phrased things better in her writing, explaining that she was writing for an academic audience and didn’t expect a wider public one. Some OpenAI executives told her that everything relating to their company makes its way into the press.
OpenAI leadership and employees were growing increasingly concerned about being painted in the press as “a bunch of effective altruists,” as one of them put it. Two days before Altman’s ouster, they were discussing these concerns on a Slack channel, which included Sutskever. One senior executive wrote that the company needed to “uplevel” its “independence”—meaning create more distance between itself and the EA movement.
OpenAI had lost three board members over the past year, most notably Reid Hoffman [who turns out to have been forced out by Altman over 'conflicts of interest', triggering the stalemate], the LinkedIn co-founder and OpenAI investor who had sold his company to Microsoft and been a key backer of the plan to create a for-profit subsidiary. Other departures were Shivon Zilis, an executive at Neuralink, and Will Hurd, a former Texas congressman. The departures left the board tipped toward academics and outsiders less loyal to Altman and his vision.
So this answers the question everyone has been asking: "what did Ilya see?" It wasn't Q*, it was OA execs letting the mask down and revealing Altman's attempt to get Toner fired was motivated by reasons he hadn't been candid about. In line with Ilya's abstract examples of what Altman was doing, Altman was telling different board members (allies like Sutskever vs enemies like Toner) different things about Toner.
This answers the "why": because it yielded a hard, screenshottable-with-receipts case of Altman manipulating the board in a difficult-to-explain-away fashion - why not just tell the board that "the EA brand is now so toxic that you need to find safety replacements without EA ties"? Why deceive and go after them one by one without replacements proposed to assure them about the mission being preserved? (This also illustrates the "why not" tell people about this incident: these were private, confidential discussions among rich powerful executives who would love to sue over disparagement or other grounds.) Previous Altman instances were either done in-person or not documented, but Altman has been so busy this year traveling and fundraising that he has had to do a lot of things via 'remote work', one might say, where conversations must be conducted on-the-digital-record. (Really, Matt Levine will love all this once he catches up.)
This also answers the "why now?" question: because Ilya saw that conversation on 15 November 2023, and not before.
This eliminates any role for Q*: sure, maybe it was an instance of lack of candor or a capabilities advance that put some pressure on the board, but unless something Q*-related also happened that day, there is no longer any explanatory role. (But since we can now date Sutskever's flip to 15 November 2023, we can answer the question of "how could the board be deceived about Q* when Sutskever would be overseeing or intimately familiar with every detail?" Because he was still acting as part of the Altman faction - he might well be telling the safety board members covertly, depending on how disaffected he became earlier on, but he wouldn't be overtly piping up about Q* in meetings or writing memos to the board about it unless Altman wanted him to. A single board member knowing != "the board candidly kept in the loop".)
This doesn't quite answer the 'why so abruptly?' question. If you don't believe that a board should remove a CEO as fast as possible when they believe the CEO has been systematically deceiving them for a year and manipulating the board composition to remove all oversight permanently, then this still doesn't directly explain why they had to move so fast. It does give one strong clue: Altman was trying to wear down Toner, but he had other options - if there was not any public scandal about the paper (which there was not, no one had even noticed it), well, there's nothing easier to manufacture for someone so well connected, as some OA executives informed Toner:
Some OpenAI executives told her that everything relating to their company makes its way into the press.
This presumably sounded like a well-intended bit of advice at the time, but takes on a different set of implications in retrospect. Amazing how journalists just keep hearing things about OA from little birds, isn't it? And they write those articles and post them online or on Twitter so quickly, too, within minutes or hours of the original tip. And Altman/Brockman would, of course, have to call an emergency last-minute board meeting to deal with this sudden crisis which, sadly, proved him right about Toner. If only the board had listened to him earlier! But they can fix it now...
Unfortunately, this piecemeal description by WSJ leaves out the larger conversational context of that Slack channel, which would probably clear up a lot. For example, the wording is consistent with them discussing how to fire just Toner, but it's also consistent with that being just the first step in purging all EA-connected board members & senior executives - did they? If they did, that would be highly alarming and justify a fast move: eg. firing people is a lot easier than unfiring them, and would force a confrontation they might lose and would wind up removing Altman even if they won. (Particularly if we do not give in to hindsight bias and remember that in the first day, everyone, including insiders, thought the firing would stick and so Altman - who had said the board should be able to fire him and personally designed OA that way - would simply go do a rival startup elsewhere.)
Emmett Shear apparently managed to insist on an independent investigation, and I expect that this Slack channel discussion will be a top priority of a genuine investigation. As Slack has regulator & big-business-friendly access controls, backups, and logs, it should be hard for them to scrub all the traces now; any independent investigation will look for deletions by the executives and draw adverse inferences.
(The piecemeal nature of the Toner revelations, where each reporter seems to be a blind man groping one part of the elephant, suggests to me that the NYT & WSJ are working from leaks based on a summary rather than the originals or a board member leaking the whole story to them. Obviously, the flip-flopped Sutskever and the execs in question, who are the only ones who would have access post-firing, are highly unlikely to be leaking private Slack channel discussions, so this information is likely coming from before the firing, so board discussions or documents, where there might be piecemeal references or quotes. But I could be wrong here. Maybe they are deliberately being cryptic to protect their source, or something, and people are just too ignorant to read between the lines. Sort of like Umbridge's speech on a grand scale.)
* note that this timeline is consistent with what Habryka says about Toner still scheduling low-priority ordinary meetings like normal just a few days before - which implies she had no idea things were about to happen.
1
Approximately four GPTs and seven years ago, OpenAI’s founders brought forth on this corporate landscape a new entity, conceived in liberty, and dedicated to the proposition that all men might live equally when AGI is created.
Now we are engaged in a great corporate war, testing whether that entity, or any entity so conceived and so dedicated, can long endure.
What matters is not theory but practice. What happens when the chips are down?
So what happened? What prompted it? What will happen now?
To a large extent, even more than usual, we do not know. We should not pretend that we know more than we do.
Rather than attempt to interpret here or barrage with an endless string of reactions and quotes, I will instead do my best to stick to a compilation of the key facts.
(Note: All times stated here are eastern by default.)
Just the Facts, Ma’am
What do we know for sure, or at least close to sure?
Here is OpenAI’s corporate structure, giving the board of the 501c3 the power to hire and fire the CEO. It is explicitly dedicated to its nonprofit mission, over and above any duties to shareholders of secondary entities. Investors were warned that there was zero obligation to ever turn a profit:
Here are the most noteworthy things we know happened, as best I can make out.
Later, when we know more, I will have many other things to say, many reactions to quote and react to. For now, everyone please do the best you can to stay sane and help the world get through this as best you can.