What happened at OpenAI over the weekend?
Board fires Sam, surprising Nadella, who hires Sam and Greg. Employees sign petition for Sam to return. Reunion still possible.
It started on Friday afternoon. I was online, and had OpenAI Twitter notifications turned on.
At first I thought the OpenAI Twitter account was hacked.
But no one was hacked.
The OpenAI board met in secret, and had resolved to dismiss CEO Sam Altman, and remove Greg Brockman as chairman of the board, as well.
This move was a surprised to Sam and Greg, who form two of OpenAI’s six board seats. People quickly concluded that the four remaining board members (OpenAI co-founder Ilya Sutskever and three independent directors) must have caucused in secret, and pushed through a board resolution without Sam and Greg.
Whether this was allowed by the company’s bylaws or not, Sam accepted his dismissal, and said goodbye on Twitter. Mind you this is on Friday, with most of the company taking the week off after OpenAI’s first Developer Day.
Greg Brockman followed up by tendering his resignation, in all lowercase, and sharing that publicly on Twitter as well.
Several other OpenAI technical leaders quit in solidarity.
It was quickly rumored that Sam and Greg would start a new company by Monday…
Fast forward to Monday afternoon — four days later.
Here is where we are now
Microsoft (OpenAI’s financial and infrastructure backer) did not know about the board room coup.
Microsoft CEO Nadella came in and tried to bring back Sam and Greg.
Sam posted a photo of himself with an OpenAI visitor badge.
Sam purportedly agreed to come back to OpenAI if the board was removed.
The board purportedly agreed to resign if they had input into the new board members.
Some progress was made, but a deadline came a went.
Before markets opened on Monday, two announcements were made
Sam Altman would join Microsoft with a CEO title, leading a new AI division within the company. Greg Brockman would follow him, as well as the OpenAI technical leaders who resigned alongside Greg.
OpenAI hired Emmett Shear, ex-CEO of Twitter and co-founder of Justin.TV, as their new “interim CEO.”
However the story may not yet be over.
500+ of OpenAI’s 700 employees signed a letter asking board to resign and for Sam to return. This number has since grown to 90% of the company.
The list of signees includes Ilya Sutskever, the OpenAI board member who either led the board’s resolution to remove SamA, or was in the room when it was decided. As well as former OpenAI Mira Murati, mentioned in the board resolution, and appointed at interim CEO by the board on Friday.
It appears that Ilya had a change of heart. And Mira may never have been in on the coup in the first place.
While Sam Altman has been announced at Microsoft, his own public messages leave it vague as whether he would possibly “reunite” the band after all. It seems that most of OpenAI’s employees would rather work for Sam, Greg and Nadella, than what has been dubbed by some as “OpenAI Classic.”
So there we are.
All in a weekend.
There’s other wrinkles, like new OpenAI Classic CEO Emmett Shear talking about investigating the board’s actions… but that’s probably of less interest to most of you.
I think this catches us up on where we are now
Sam is CEO of a Microsoft division, with full backing from Nadella, and inheriting most of OpenAI’s technical leadership.
OpenAI will continue to function — for the time being — with commitments to keep the API and ChatGPT running. Probably with technical support from Microsoft and ex-OpenAI people, even if they shift over to Sam’s new company.
Microsoft stock is up… 2%!
Sam, Greg and Nadella may yet get the band back together…
To this I’ll add a couple of my own thoughts and predictions
ChatGPT and the API will remain operational. Unless there’s a massive uptick in traffic… these systems are replicable — I doubt that Microsoft won’t help OpenAI keep the lights on
Training GPT5, or any new models…. is another matter
The GPTs / GPT Store that just launched, was Sam’s thing and many research / safety minded people were critical of it
I assume that in the short run, things at OpenAI will operate more smoothly that doomers expect. But in the medium term, progress at OpenAI will come to a halt, at least as they reorganize and many people leave. In the long run, I don’t see them supporting something as product and commercially minded as the GPT Store — or perhaps even the GPT API and ChatGPT.
Before this whole thing started, weeks ago Ben Thompson suggested that OpenAI should discontinue the GPT API and let Microsoft handle that. He suggested that they should focus on researching cutting edge models, and on a singular “ChatGPT” product — possibly with its own app store.
This may be exactly what happens. Except that the ChatGPT product will be part of Microsoft under Sam’s new division.
OpenAI Classic may continue to research bigger and more intelligent models. Or it may deprecate itself, as talent and resources flow to the new Sam Corp inside Microsoft, and transitional help to keep OpenAI running dries up.
There are now half a dozen APIs you can use to get LLMs at least as good as GPT3.5, and some comparable with GPT4… not overall but on specific tasks.
Now we know what’s possible, and AI progress doesn’t sleep.
We will see what happens, but this is probably a good day for “decentralized” AI as Balaji calls it, for open source AI and LLMs, and maybe a good day for all of us trying to accelerate into the future.
PS I may add more later on the details of how we got here. About EA vs e/acc… but perhaps a story for another time.