Why Did Sam Altman Get Fired?

silly political shit or AI safety squabbles

Sam Altman got fired from OpenAI. They made a statement which is the usual PR-speak but this fragment stood out:

“Not consistently candid” to me screams that there is a lot more to this story —performance and poor image are obviously out of the question, so the issue must be related to politics or ethics. Some people have speculated that it may be related to sexual harassment:

The psychological characteristics that predispose one to rape are probably stable, so if he did do it, then it’s possible that he has done more rapey stuff at his own company. While I originally thought this was somewhat plausible, I think it can be rejected after this tweet:

The exit tweet is too positive, and if he was canned for raping people, he probably wouldn’t have much to say about “what’s next later”.

This leaves three possibilities:

1. He got outsed because on an internal battle over AI safety.

2. He has bigger plans, and they fired him to attract attention.

3. He got outsed because of a silly political squabble.

Before I make the case for any particular theory, I think it’s worth analyzing the current board of directors. We have:

• Greg Brockman: the chairman of the board who just stepped down (????). He has tweeted about AI safety 14 times, most of the time endorsing a lukewarm pro-safety take.

• Ilya Sutskever: tech/AI researcher. based on a quick google search, seems very pro-safety.

• Adam D’Angelo: CEO of Quora. Latecomer to the board of directors, has come out as pro-safety once in an interview and a quora post.

• Tasha McCauley: not much about her online. An effective altruist (a zealot who commits their life towards “altriusm” and is associated with AI safety, acronym is EA), according to people closer to this community than I am.

• Helen Toner: director of some think tank that does tech/security research, which includes some AI safety research. Also EA adjacent, apparently.

The fact Greg stepped down suggests that there is some inter board member conflict as well, probably between Greg and the rest of the board. This makes theory 2 much less likely to be true, as Sam Altman departing for other opporitunities is unlikely to have caused animosity between the board members. It’s also worth mentioning that none of these members have tweeted about the incident as of 12:55 AM CET, about four hours after the original announcement.

Given that the board is generally pro safety, this could be a potential source of disagreement. Sam Altman has been fairly even-handed about AI safety as an issue, arguing that regulation is necessary, but that current LLMs do not warrant any regulation. This is the safest public position to take - going to hard for safety would hamper OpenAI’s success, while flouting a commitment to it would make the regulators and safetyists wary.

Scott Alexander noted that if OpenAI was actually concerned about AI safety, they would make more tangible commitments against burning timeline - which I am in agreement with. There are obviously a lot of nuances involved with this argument - I would recommend reading Scott’s article if you want to dive into them.

Theory 3 (political squabble) is still a possibility, of course. Sam Altman’s influence stems not only from his title but also from his substantial social media presence and the loyalty he commands among the staff - which is a power threat to the board. If the board were to depose Sam and install their own CEO, their power would increase greatly. Among the five board members, Ilya and Greg seem to have closer ties to Sam, while the others appear to be more loosely connected to him. However, I am not the best judge of this - I am only marginally linked to the AI/tech community.

All in all, the most likely possibilities at the moment appear to be that he got outsed in a political squabble or over an ideological conflict over AI safety. The current evidence does not support that he has done anything particularly heinous, and he probably was not outsed to help him jumpstart his own grift. I previously joked that it could be because he has an alt where he posts sonnenrad edits and yaoi - questionable political beliefs/allies are probably not at stake here, but I don’t think it’s totally out of the question. Tech twitter is only a few nodes away from right wing twitter - there are a lot of shared ties between the two communities like Zero HP Lovecraft or Lukas.

Updates:

An alternative theory suggests that this is an OP from the US govt to oust an upcoming powerful actor and replace them with a govt-friendly leader. A bit far fetched, but far from impossible.

Alternative theory on HN states it was due to a data breach. A bit much to fire a CEO for this, but doesn’t seem that implausible.

New tweet suggests to me that the safety theory is probably true. Though, that doesn’t rule out the possibility that there was a political component to the coup as well - especially when ideological divides often reflect political ones.

Looks like my intuition that this was a political/ideological conflict was correct. Confirmation from Roko and Shkreli: