[recovery_mode]

'OpenAI’s "primary fiduciary duty is to humanity," not to investors or even employees.'

I'm so proud of the board for sticking to their principles over profit. This is a huge victory for all of us, and a chance for everyone to come together and approach AGI safely and for the benefit of all.

[paulddraper]

I'm not a believer in "AI safety."

But if you do believe, I don't see how this a "victory."

OpenAI is now existentially at risk. Even if it does survive it will very likely lose its dominance to a non-"AI safety" commercial endeavor.

---

Classic "biting the hand that feeds you" https://youtu.be/Lg2dqFCU67Q?si=TWX8slGW_8hLdgQu&t=35

[yeck]

By that do you mean that you believe AI is safe by default? Or that AI safety is not possible?

[paulddraper]

I mean that "AI Safety" is (1) nebulous and (2) impossible.

Encryption can be used for good things and bad things. And yet no one is talking about "encryption safety."

[yeck]

> And yet no one is talking about "encryption safety

Okay, you are getting me off topic, but this is totally not the case. Hasn't governments been especially interesting in dealing with encryption lately? UK or EU (can't remember which) being most recently raising the pitchforks.

Anyway, the more important thing for me to point out is that AI is anticipated to possess exponential properties that technology like encryption does not have. Encryption is predictable while AI is not.

Edit: fixed typo

[paulddraper]

> Hasn't governments been especially interesting in dealing with encryption lately?

Fair enough :)

People have been talking about it, and it's a similarly bad idea.

> Encryption is predictable while AI is not.

Banking info vs child porn is pretty divergent.

[yeck]

They are both just data being made hard to read, though. You could pretty much sum up the "risks" of encryption as: you want to read some data that someone else didn't want you to read. Which means you might fail to detect something, or fail to prove something as easily as if you could read that data. While we could enumerate specific examples of scenarios that fit the generalization, I think we are talking about something pretty narrow. It is also not really a new problem. Historically communication has been impossible to snoop on at scale, and encryption exists to maintain a semblance of past privacy in the present.

Lol not sure where I want to take this and it's getting pretty late...

[snewman]

The board's motivation may have been good, they may be "in the right" here (it depends on your view on AI risk, and also on a lot of details of what was going on that we still don't know). But unambiguously they executed abysmally. There is little to be proud of here. OpenAI may have needed to be reined in, and they have probably succeeded in accomplishing that, but they have badly damaged the company in the process, and catastrophically damaged their own credibility, thus reducing their ability to influence events in the future and plausibly damaging the credibility of the entire idea of responsible development. (To be clear, I believe responsible development is important and I am sad to see it discredited.)

Edit to add: if you believe the rapid commercialization of frontier models at OpenAI was a cancer that needed to be destroyed, the likely outcome of the board's clumsy handling is that it probably now metastasizes.

[azinman2]

You're assuming it'll continue on that path, and well.

Personally I think this is where they'll end up stagnating, although I'm happy to be wrong. It's going to make recruiting top talent far harder, and ultimately this hiccup will cause them to lose their edge. Losing that edge lessens the desire to go there, which will drive down competitiveness, which will further pull them back from the fray. I believe it'll make the US less competitive, which makes me sad.

[yeck]

Hard to say what the impact on access to talent will be. I know people working in AI that wouldn't consider working at OpenAI out of perceived recklessness. At least one that I've been speaking with during this weekend said that this made them more open to the idea.

[Racing0461]

> approach AGI safely

The only risk AGI poses to the average american is unemployment. If immigration/outsourcing is fine, i don't see anthing being done about this.

[reducesuffering]

Sam Altman, Ilya Sutskever, Yoshua Bengio, Geoff Hinton, Demis Hassabis (DeepMind CEO), Dario Amodei (Anthropic CEO), and Bill Gates disagree with you.

https://twitter.com/robbensinger/status/1726039794197872939

[Racing0461]

They only signed that to pull the ladder up after they have a mass adopted model to lower competition. Nothing about that is about concern for humanity, it's about lowering competition to strengthen their moat.

Now that sam is out and his new ai startup if he makes one will be a new entrant in the ai space, he probably should have waited a couple weeks before recommending all the ai regulations.

As an aside, i would support a system like the patent system where if new ai models are going to be heavily regulated, then the uncensored/pre-gated weights of the already launched model should be released to the public in much the same way your patent idea is protected but you have to release the design in the application.

[reducesuffering]

No, Sam Altman and others have written about this a long time ago. 8 years ago:

"Why You Should Fear Machine Intelligence"

https://blog.samaltman.com/machine-intelligence-part-1

It's unfortunate how crank people on HN like you turn when it comes to this grave topic. Multiple people, for years, many without incentive, will tell you what they think and you will just "no they're lying," especially when YC has placed a huge amount of value on earnestness forever. Not too mention they've addressed the regulation conspiracy and how the regulation restricts them and big co's, not GPT4 level models by anyone smaller.

[Racing0461]

May i interest you in a bridge, sir?