Satya Nadella says OpenAI governance needs to change
cnbc.com
cnbc.com
Incredible. Anyone with FTC knowledge know if making those sorts of announcements can run afoul of regulations?
Edit: Sorry, SEC.
They had that last week. If Sam returns to OpenAI, they will continue to have it. If Sam brings most of the OpenAI team to MS, there will be an awkward recreation of IP and MS will likely get it again [0]. But other than being the one to bring the team along, I’m not sure what Sam, personally, would bring.
(Sam at OpenAI is a different story obviously.)
[0] OpenAI has some maybe-dubious IP in the form of a massive dataset. MS may prefer not to possess such a thing. And I don’t know whether MS is permitted under its contract to create derivative works of whatever it licensed from OpenAI.
Sam Altman replacing Satya Nadella as CEO of Microsoft would certainly be a twist!
Sam Altman was never a consideration as CEO of the whole company.
But instead CEO or VP of a seperate entity similar to how Github operates for which Sam would be more than suitable. And once again Sam just built a $100b business. That requires a bit more skill than people give him credit for.
Building a $100bn business is one thing. Running a division of a big company is rather different.
But beyond that did Nadella think there was insufficient chaos around OpenAI and felt he needed to add his own?
Everyone is issuing claims, counterclaims and threats. Has no one thought it might be a good idea to just step back and think for a day or two, let things cool down before issuing more of the same?
That would be a public admission of securities fraud, with an admission that they did so intenteding to commit securities fraud.
Confessing securities fraud, or
Doing more securities fraud.
As Matt Levine likes to say - everything is a security's fraud and in this case, it might actually be. It is now abundantly clear that those statements were merely to hold the stock price drop come Monday.
Lying about what specifically?
And remember, he did this AFTER the markets closed, so he was sitting on this information the whole day.
This was the tweet. Read it carefully.
Here's a bit more context -
You think that Satya would say “Sam is joining” if Sam didn’t at least verbally accept an offer?
You win for stupidest comment on the internet.
[0] For the record, AI safety is the branch of research worried about controlling computer God, while AI ethics is the branch of research worried about whether or not it's OK to pay Pakistani kids pennies to label training data.
Would you also include in “AI ethics” a considered rollout of automation to maximize its benefit to everyone? Because there is real benefit. The kind of work AI can automate is shit work and it would be to the theoretical benefit of society for it to be automated… unless all the benefits accrued to the owners of the tech instead of the people whose (stupid) jobs are being replaced.
Is that “AI ethics” or still a third (anti capitalist?) thing?
The workers doing the jobs will never benefit. They'll be shown the door and the owners will reap all the profits.
There is definitely that second and more abstract take - AI doomsday safety - that Altman liked to talk about to get favorable regulation passed. But it doesn't seem to be much of a real field. "Hard" AI doomers think that superhuman AI can't be reliably constrained and we should just stop. "Soft" doomers think we need to solve "alignment". But we don't really know what's going on inside an LLM, so we test for alignment by asking it what it thinks about Hitler. In other words, back to brand safety.
As a key example, we were the first team to discover Prompt Injection, which we responsibly disclosed to OpenAI leadership in early May of 2022, four months before it was independently rediscovered by Riley Goodside.
(See the citation at the bottom of this page https://github.com/OWASP/www-project-top-10-for-large-langua... )
Prompt Injection is a perfect example of what we consider a near-term “AI safety” issue: it’s a real issue that affects current systems, and it will also affect any instruction-based AGI systems if a solution is not found by the time AGI is reached. There is currently no known general solution to Prompt Injection, and it’s #1 on OWASP’s recent list of the top vulns in LLMs. (See link above. This vuln is designated LLM01.)
Also, the two are tied: those underpaid, overworked Filipinas? Surprisingly reactionaries.
“Wolf thinks sheep shouldn’t be cooped up in secure pen”.
Still no Form RRF-1 filed today.
They are Open AIs largest investor and gives Open AI a steep discount on compute.
Microsoft already has a perpetual license to all of OpenAIs IP and could easily hire every single one of thier employees without blinking
https://www.wsj.com/articles/microsoft-and-openai-forge-awkw...
Think of what Apple can do with a license to ARM and now compare that to what MS could do with a license to GPT 3?
Do you really think any company would depend on another companies technology the way that MS is dependent on OpenAI without having a favorable license?
I’ve dealt with something similar on a very minute scale on both sides - being a dev lead at a startup where our largest customer accounted for 70% of our revenue and being the dev lead where the company I was working for accounted for 70% of a vendors revenue.
Each time, I saw how much power the customer yielded. It involves lawyers, licensing, contingencies, etc. and in both of the cases above - escrowing of code where the customer got the rights in certain conditions.
I’m 100% sure that MS has better lawyers than I encountered.
You can't compare a hostile OpenAI with a cooperative ARM. Even with a cooperative OpenAI, Microsoft restricts internal researcher access to the source code, so the license appears less favorable than you think.
Again I’m 100% sure that Microsoft wouldn’t base its entire technology line on another company’s IP without having very favorable licenses. That’s just not how these things work.
Licensing ChatGPT versions developed by OpenAI seems favorable enough to me. The question boils down to what happens if OpenAI stops development. I highly doubt the bright minds at OpenAI gave away their IP for widely available compute credits.
Unless they are a gang of thieves, they are all exchanging labor for money to support their addiction. Who do you really think has the negotiating power in this relationship?
The smaller OpenAI that signed the Microsoft agreement had plenty of sustenance. Compute remains fungible unlike LLMs.
And if you think a migration at that scale is “easy” you haven’t been involved with mass migrations. Besides, Microsoft’s enterprise customers aren’t going to jump ship to another cloud provider and follow OpenAI.
Before I was giving anecdotes from n-6 job working at a startup and my n-3 job when I was on the other side of the table as a vendor’s largest customer.
This part of knowing how the cloud industry and migrations work comes from my n-1 job working in the Professional Services Department at AWS when I was involved with large migrations and saw how sticky enterprises were when it came to dealing with Microsoft.
And what does OpenAI have to offer another company with Microsoft already having 49% equity?
Second, I meant the OpenAI that signed the initial agreement with Microsoft years ago already had ample resources and funding. In those negotiations, OpenAI would have had little incentive to hand over all their IP regardless of the offer on hand.
Also, it’s standard in many contracts that you can’t actively poach a partner’s employees. That doesn’t mean that if the employees actively reach out to you, you can’t hire them. I saw that at AWS.
If they had so much funding why were they willing to give up 49% of the company to Microsoft? Does that read like a company that could be picky?
OpenAI gave up 49% equity because they thought Microsoft gave a fair valuation and terms, among other possible reasons. OpenAI wanted to scale up their LLM research so they signed the agreement. I don't think OpenAI negotiated from a position of weakness.
The only strategic play left to OpenAI is spite. They can choose to burn down Chat GPT which will lead to a scattering of their talent across competitors (although Microsoft will likely just pick up most). This will slow down Microsoft by a couple of months but will be nothing more than a bump on their way to AI dominance.
The board should have led with this as an implied threat behind closed doors and given both Altman and Nadella a face-saving public retreat. Altman goes off to start GPTSoft with equity, a license to GPT4 and access to Microsoft's cloud compute. Everyone gets to call this a win. Public humiliation was a mistake.
That's not a problem for the non-profit board. If the mission is best served by the non-profit ceasing to exist, they should make it cease to exist.
Why he thinks that he has authority to dictate how the non profit should be structured? I don't think he has.
https://assets.bwbx.io/images/users/iqjWHBFdfxIU/ixMQ90DVObm...
Why he thinks that he has authority to dictate how the non profit should be structured? I don't think he has.
They have no authority on paper. They have plenty of authority in practice by virtue of exercising their power/leverage inherent in OpenAI’s dependence on them for compute resources.
Because, as Matt Levine pointed out today, there are two ways that you can view OpenAI's corporate governance.
On the one hand, you have the board, which has legal control over the entire company and all of its subsidiaries.
On the other hand, you have Microsoft, who has money.
Under the world view of contracts and corporate governance structures, OpenAI is the company of the board. The board can do whatever they want.
Under the world view of 'people like to eat sometime this month, and OpenAI needs to fundraise', OpenAI is a subservient division of Microsoft. Who can do whatever it wants, up to, and including acquiring all of its IP and staff for essentially $0, with zero regulatory oversight or anti-trust problems.
On paper, you may have property rights to your house. In practice, someone who has property rights around your house can build a wall around it, and wait for you to either sell it to him for a dollar, or starve.
This whole adventure has definitely been a novel (if expensive) take on the 3E's of Microsoft - 'Embrace, Extend, Exterminate'.