News for hackers. Stuff that matters.

A comment on OpenAI's employees were given two explanations for why Sam Altman was fired. Parent comment

Re:OpenAI's employees were given two explanations for why Sa... by hal009 on Monday November 20, 2023 @10:16PM (#38358623)
As mentioned by another person in this thread [0], it is likely that it was Ilya's work that was getting replicated by another "secret" team, and the "different opinions on the same person" was Sam's opinions of Ilya. Perhaps Sam saw him as an unstable element and a single point of failure in the company, and wanted to make sure that OpenAI would be able to continue without Ilya?

[0] https://news.ycombinator.com/reply?id=38357843

[ Parent | on HN ]
Re:OpenAI's employees were given two explanations for why Sa... by JimDabell on Monday November 20, 2023 @11:03PM (#38359014)
Since a lot of the board’s responsibilities are tied to capabilities of the platform, it’s possible that Altman asked for Ilya to determine the capabilities, didn’t like the answer, then got somebody else to give the “right” answer, which he presented to the board. A simple dual-track project shouldn’t be a problem, but this kind of thing would be seen as dishonesty by the board.
[ Parent | on HN ]
Re:OpenAI's employees were given two explanations for why Sa... by hn_throwaway_99 on Tuesday November 21, 2023 @01:30AM (#38360120)
> it’s possible that Altman asked for Ilya to determine the capabilities, didn’t like the answer, then got somebody else to give the “right” answer, which he presented to the board.

This makes no sense given that Ilya is on the board.

[ Parent | on HN ]
Re:OpenAI's employees were given two explanations for why Sa... by JimDabell on Tuesday November 21, 2023 @01:56AM (#38360294)
No, it just means that in that scenario Sam would think he could convince the rest of the board that Ilya was wrong because he could find somebody else to give him a preferable answer.

It’s just speculation, anyway. There isn’t really anything I’ve heard that isn’t contradicted by the evidence, so it’s likely at least one thing “known” by the public isn’t actually true.

[ Parent | on HN ]
Re:OpenAI's employees were given two explanations for why Sa... by kmlevitt on Monday November 20, 2023 @10:46PM (#38358874)
Firing Sam as a way of sticking up for Ilya would make more sense if Ilya wasn’t currently in support of Sam getting his job back.
[ Parent | on HN ]
Re:OpenAI's employees were given two explanations for why Sa... by onimishra on Tuesday November 21, 2023 @03:09AM (#38360801)
I’m not sure Ilya was anticipating this to more or less break OpenAI as a company. Ilya is all about the work they do, and might not have anticipated that this would turn the entire company against him and the rest of the board. And so, he is in support of Sam coming back, if that means that they can get back to the work at hand.
[ Parent | on HN ]
Re:OpenAI's employees were given two explanations for why Sa... by kmlevitt on Tuesday November 21, 2023 @04:17AM (#38361291)
Perhaps. But if the board is really so responsive to Ilya's concerns, why have they not reversed the decision so that Ilya can get his wish?
[ Parent | on HN ]
Re:OpenAI's employees were given two explanations for why Sa... by 015a on Monday November 20, 2023 @10:50PM (#38358911)
This is an interesting theory when combined with this tweet from Google DeepMind's team lead of Scalable Alignment [1].

[1] https://twitter.com/geoffreyirving/status/172675427761849141...

The "Sam is actually a psychopath that has managed to swindle his way into everyone liking him, and Ilya has grave ethical concerns about that kind of person leading a company seeking AGI, but he can't out him publicly because so many people are hypnotized by him" theory is definitely a new, interesting one; there has been literally no moment in the past three days where I could have predicted the next turn this would take.

[ Parent | on HN ]
Re:OpenAI's employees were given two explanations for why Sa... by doktrin on Monday November 20, 2023 @11:12PM (#38359103)
Interesting. I’m glad he shared his perspective despite the ambiguity.
[ Parent | on HN ]
Re:OpenAI's employees were given two explanations for why Sa... by nvm0n2 on Tuesday November 21, 2023 @09:46AM (#38364301)
That guy is another AI doomer though, and those people all seem to be quite slippery themselves. Supposedly Sam lied to him about other people, but there's no further detail provided and nobody seems willing to get concrete about any time Altman has been specifically dishonest. When the doomer board made similar allegations it seemed serious for a day, and then evaporated.

Meanwhile the Google AI folks have a long track record of making very misleading statements in public. I remember before Altman came along and made their models available to all, Google was fond of responding to any OpenAI blog post by claiming they had the same tech but way better, they just weren't releasing it because it was so amazing it just wasn't "safe" enough to do so yet. Then ChatGPT called their bluff and we discovered that in reality they were way behind and apparently unable to catch up, also, there were no actual safety problems and it was fine to let everyone use even relatively unconditioned models.

So this Geoffrey guy might be right but if Altman was really such a systematic liar, why would his employees be so loyal? And why is it only AI doomers who make this allegation? Maybe Altman "lied" to them by claiming key people were just as doomerist as those guys, and when they found out it wasn't true they wailed?

[ Parent | on HN ]
Re:OpenAI's employees were given two explanations for why Sa... by BillyTheKing on Monday November 20, 2023 @11:06PM (#38359044)
either that or Sam didn't tell Adam D'Angelo that they were launching a competing product in exactly the same space that poe.ai had launched one. For some context, poe had launched something similar to those custom GPTs with creator revenue sharing etc. just 4 weeks prior to dev-day
[ Parent | on HN ]
Re:OpenAI's employees were given two explanations for why Sa... by m3kw9 on Tuesday November 21, 2023 @12:16PM (#38366647)
Not sure how he would see that coming? It was a UI tweak away for OpenAI
[ Parent | on HN ]