Hacker News Reader

Comment by mfiguiere | Hacker News Reader

TheInformation: Dozens of Staffers Quit OpenAI After Sutskever Says Altman Won’t Return

>Dozens of OpenAI staffers internally announced they were quitting the company Sunday night, said a person with knowledge of the situation, after board director and chief scientist Ilya Sutskever told employees that fired CEO Sam Altman would not return.

https://www.theinformation.com/articles/dozens-of-staffers-q...

View on HN

Isn't this expected? Nearly everyone who joined post ChatGPT was primarily financially motivated. What is more interesting is how many of the core research team stays.

This is actually pretty surprising to me, since a financially motivated person would normally wait until a better deal, and just collect their paycheck in the meantime.

There's also no guarantee that Altman will really start a new company, or be able to collect funding to hire everyone quickly. I wonder if these people are just very loyal to Sam.

> This is actually pretty surprising to me, since a financially motivated person would normally wait until a better deal, and just collect their paycheck in the meantime.

I imagine you need to signal that you want in on the deal by departing. Get founder equity.

Even if he had started a new company, there was no way a dozen employees were getting founder equity for showing loyalty

Or they could be loyal to the e/acc cult.

How do you know that? Maybe they wanted to ship AI products at an unprecedented speed at the most prestigious AI company in the world.

This. Very accurate. At the end of they day this is a battle between academics and capitalists and what they stand for. We generally know how this typically goes…

I don't see many academics indulge in sensationalist doomsaying. That's the real difference here. SETI wouldn't and couldn't seek grants by proposing to contact murderous aliens.

I think academics have a general faith in goodwill of intelligence.Benevolence may be a convergent phenomenon. Maybe the mechanisms of reason themselves require empathy and goodness

Huh? There's plenty of AI doomerism amongst academics, see Bengio, Hinton, etc...

Hinton makes cliched statements as if he's not given much thought to safety but feels obliged for whatever reason

The capitalists run it into the ground while the academics stand around confused asking each other what happened?

Tip for builders: you can use the GPT APIs on Microsoft Azure. Managed reliably, nobody's quitting, no drama. Same APIs, just with better controls, global availability, and a very stable, reliable, and trustworthy provider. (disclosure: I work at Azure, but this is just my own observation).

GPT on Azure has become incredibly slow for us in the past few weeks.

The Azure-hosted versions are consistently behind the OpenAI versions.

For example, the GPT4 128K-token model is unavailable, and the GPT-4V model is also unavailable.

This. Very frustrating. Why is Azure behind and when is the gpt-4-turbo version coming?

It's already available globally on Azure as of last week.

Okay you’re correct. Last week when I checked I only saw the Dall-E 3 public preview announcement. Now I checked and the Azure page is updated also with GPT-4 Turbo announcement. Very nice!

They only have the 8K token version.

You want me to trust M$ in all this? Embrace, extend, extinguish.

Fellow nerds, you really need to go into work on Monday and have a hard chat with your C levels and legal (Because IANAL). The question is: Who owns the output of LLM/AI/ML tooling?

I will give you a hint, it's not you.

Do you need to copyright what a CS agent says, no, you want them on script as much as possible. An LLM parroting your training data is a good thing (assuming a human wrote it). Do you want an LLM writing code, or copy for your product, or a song for your next corporate sing along (Where did you go old IBM)? No you dont, because it's likely going straight to the public domain. Depending on what your doing with the tool and how your using it, it might not matter that this is the case (its an internal thing) but M$, or openAI, or whoever your vendor is, having a copy that they are free to use might be very bad...

Have I just been transported to Slashdot in 2003?

I'm not sure you appreciate how enterprise licence agreements work. Every detail of who owns what will have been spelled out, along with the copyright indemnities for the output.

a hint - the "M$" thing is not smart or funny, just old.

Also, you might be given someone else’s proprietary IP, setting yourself up for a lawsuit.

If I grab something off GitHub, and the license there is GPL, but it was someone else's IP I do have some recourse and for my infraction.

In the case of an LLM handing it to me can I sue MS or OpenAI for giving out that IP, or is it on me for not checking first? Is any of this covered in the TOS?

> Embrace, extend, extinguish.

Microsoft hasn't embraced that ideology in close to more than a decade by now. Might be the time to let go of the boomer compulsion.

I don't understand what point you're trying to make. Yes Microsoft uses OpenAI APIs. What is the point you're trying to make beyond that? It's still OpenAI software.

Yes, the model weights were developed by OpenAI. They are licensed exclusively and irrevocably to Microsoft, and operated by Microsoft, not OpenAI. If you are building with these APIs and concerned that consuming them from OpenAI (which also runs them on Azure, but managed by OpenAI staff) because of the drama there, you can de-risk by consuming from Azure directly.

> They are licensed exclusively and irrevocably to Microsoft, and operated by Microsoft, not OpenAI.

No wonder why CEO got fired.

Let me guess: Ilya and his team had developed GPT5, decided it very-nearly had consciousness, and then Sam immediately turned around and asked Microsoft what they're willing to pay for a copy to use and abuse.

If folks care enough to move to Azure, I think they might as well derisk entirely from OpenAI models, despite its quality?

Microsoft doesn't "use" the APIs, they host them on their own servers and have a license to do so and re-license to Azure users. If something goes wrong with OpenAI (given that it sounds like many key employees are leaving), Azure will stay up and you can keep using the APIs from MS.

Or, to say it another way, they are cooperating with OpenAI - OpenAI uses Microsoft's cloud services, and Microsoft incorporates OpenAI's products in its own offerings. But the worries people have are not about OpenAI's products suddenly vanishing, it's about the turmoil at OpenAI affecting the future of those products.

Actually the exodus of talent from OpenAI may turn out to be beneficial for the development of AI by increasing competition - however it will certainly go against the stated goal of the board for firing Altman, which was basically keeping the development under control.

That may provide short term stability, but medium term (which in this field is a few months) how will Azure's offering move forward if OpenAI is in such crisis? I guess it really comes down to OpenAI's ability to continue without Altman and Co. I don't believe that Microsoft's license allows them to independently develop the models? Wouldn't this become a stale fork pretty quickly while the rest of the industry moves on (llama2 etc ..)?

Or ... cut the middleman: Sam Altman and Greg Brockman joining MS to start a new AI unit - https://twitter.com/satyanadella/status/1726516824597258569

I agree that medium term is up in the air and highly dependent on what happens next. If many OAI employees defect to Sam's new company, maybe that becomes the thing everyone migrates to...

The current models would presumably be accessible for customers regardless of OpenAI’s state. If OpenAI were to hypothetically somehow vanish into thin air, products and features built on their products could still be supported by Azure’s offering.

Sure, but what's the point on building a product on top of a stable API that exposes a technology that won't evolve because it's actual creators have imploded? It remains to be seen whether OpenAI will implode, but at this point it seems the dream team is t getting back together.

Question: how difficult is it to get that no retention waiver on prompts and responses?

Not difficult. I've not heard of anyone who asked and _didn't_ get the waiver. It's just a responsible stop-gap in case a user does something questionable or dangerous.

The waiver still allows for logging of prompts for the specific purpose of abuse monitoring for some limited retention period, right? How difficult is it to have this waived as well?

I work in academia and with somewhat protected data so YMMV but it wasn't hard for me at all (I just filled out the form and MS approved it).

How the same? Does it have the new Assistant API too?

Basically, yes (there are some variations but same functionality, and much more).

Does anyone have a non-paywall version of this? Or like excerpts from the article?

The information is a 300 dollar annual subscription, I don’t think they will allow it

Oh wow, that's like the most I've seen for any news subscription.

https://archive.ph/Berx9

Reply on news.ycombinator.com