921–930 of 1001 posts
Earlier quoted context omitted.
See The Linux Foundation, they don’t seem to have this problem.
I used to work for an opensource organization full time. They used to rely on contracting for revenue. It wasnt a lot but it was sufficient to keep the org going for 10 years.
It's clear that OpenAI has become something that it wasn't intended to be at it's founding. Maybe that change happened for good reasons, but the fact that there was a change is not in doubt.
Intention is an interesting word. I wonder how many of the founders quietly hoped it would make them a lot of money. Though to be fair, I do believe that hope would have been tied to the expectation that they meet their stated goals of developing some form of AGI.
Earlier quoted context omitted.
This only holds if defense outscales attack. It seems very likely that attack outscales defense to me with LLMs.
Well then, isn’t the whole case about just denying the inevitable? If OpenAI can do it, I would not say that that is very unlikely for someone else to do the same. Open or not. The best chance is still that we prepare with the best available information.
Earlier quoted context omitted.
>Surely GPT-4 would fail some exams Some? It does hilariously badly on basic math. With confidence, though.
GPT-4 with code interpreter is better at math than elite Math undergrads.
Earlier quoted context omitted.
> Not seeing Mistral Large as an open source model now has a bitter taste to it. A company needs a product to sell. If they give away everything, they have nothing to sell. This was surely always the plan. (1) They can give away the model but sell an API - but they can’t serve a model as cheap as Goog/Msft/Amzn who have better unit economics on their cloud and better pricing on GPUs (plus custom inference chips). (2)…
> (1) They can give away the model but sell an API - but they can’t serve a model as cheap as Goog/Msft/Amzn who have better unit economics on their cloud and better pricing on GPUs (plus custom inference chips). Which has a simple solution, release the model weights with a license which doesn't let anyone to commercially host them (like AGPL-ish) without your permission. That is what Stability.ai does it.
Earlier quoted context omitted.
Intention is an interesting word. I wonder how many of the founders quietly hoped it would make them a lot of money. Though to be fair, I do believe that hope would have been tied to the expectation that they meet their stated goals of developing some form of AGI.
It seems a bit weird to quietly hope that the money you put in an organization with the explicit goal of being a non-profit, would give you direct monetary returns though.. Maybe they hoped for returns in other ways, like getting some back-channel AGI love when it finally became conscious? :)
So while the non profit is a specific legal entity that has a certain clear function, one may still want to use this public benefit open source project for for-profit means.
However this doesn’t really apply to OpenAI, because their system is not open source and also because strangely, their non profit owns the for-profit. Non-profit founders could theoretically be fine desiring profit, but the way OpenAI has done it seems particularly strange.
Earlier quoted context omitted.
> Not seeing Mistral Large as an open source model now has a bitter taste to it. A company needs a product to sell. If they give away everything, they have nothing to sell. This was surely always the plan. (1) They can give away the model but sell an API - but they can’t serve a model as cheap as Goog/Msft/Amzn who have better unit economics on their cloud and better pricing on GPUs (plus custom inference chips). (2)…
> (1) They can give away the model but sell an API - but they can’t serve a model as cheap as Goog/Msft/Amzn who have better unit economics on their cloud and better pricing on GPUs (plus custom inference chips). Which has a simple solution, release the model weights with a license which doesn't let anyone to commercially host them (like AGPL-ish) without your permission. That is what Stability.ai does it.
Earlier quoted context omitted.
Are you a lawyer or have some sort of credentials to be able to make that statement? I’m not sure if Elon Musk being hypocrite about AI safety would be relevant to the disputed terms of a contract.
I don't think it's about him being a hypocrite - just him undermining his own argument. It's a tough sell saying AI is unsafe but it's still in the public's best interest to open source it (and hence OpenAI is reneging on it's charter).
This is, in many ways, the substance of the lawsuit. This logic of "we must guard this secret carefully... for safety!" doesn't actually inevitably come from most lines of enabling research in any field in academia for example, but it does reliably come up once someone can enclose the findings in order to profit from exploiting this information asymmetry somehow
Secrecy for profit isn't a super benevolent thing to do, but it's generally speaking fine. We have whole areas of law about how to balance the public benefit of wide availability of information and the private benefit to discoverers of some technique, technology, or even facts about the world. It is well understood by most people that trade secrets aren't public knowledge. We see this plea to "safety" come up only exactly in cases where companies want to justify having control over things that have become pervasive and often mandatory to use in many contexts in a way that allows said companies to in turn exert further control over that thing's users, which is to say in tech monopolies. The use of that reasoning basically one-to-one predicts a business model that relies on DMCA 1201 (or its international equivalents) to function, a legal edifice designed by Microsoft lawyers which has become pervasive worldwide essentially at their behest
That said, I don't think it's particularly hard to make the case that writing a whole-ass non-profit charter explicitly outlining the intent to do research in the open and then suddenly switching to the very familiar corporate reality distortion field stance of a convicted monopolist you happen to have formed a partnership with in order to justify effectively abandoning that charter is a good basis for a lawsuit
Earlier quoted context omitted.
In a company (it may be complicated due to OpenAIs structure) the boards sole purpose is to represent all shareholders. If they don’t that’s usually asking for a SEC investigation or private law suit.
Yes, if we just ignore OpenAI's unusual structure it really simplifies the discussion, much like the joke about the physicist who starts by assuming a perfectly spherical cow.
No it only makes who the responsibility of the board is to - the non profits charter, the donors, or the shareholders. That decides if their decision was lawful, not their intent.
Earlier quoted context omitted.
>Surely GPT-4 would fail some exams Some? It does hilariously badly on basic math. With confidence, though.
Have you tried GPT recently on maths? Since they trained it to write code for maths questions it's got a lot better