Software is always more important than hardware. All the big players have access to NVIDIA chips today and yet only OpenAI has ChatGPT, proving the point.
OpenAI probably wishes someone would create competition to NVIDIA and this is Sam Altman trying to make that happen himself, since no one else seems to have been able to pull it off so far.
A conflict of interest would be OpenAI buying Altman's chips at inflated prices or something like that.
But if he makes a bunch of money selling OpenAI chips and OpenAI gets better/cheaper chips, that seems like pure win-win and totally free of ethical conflict.
And in the tech-specific case: As much as junior engineers would love to believe in "superior" solutions, tech decisions are seldom clear-cut. There are many trade-offs: cost, efficiency, memory use, throughput, latency, ease of use, cost of switching, and many more. You always have a pile of pros and cons. Sometimes, one is strong enough, but most of the time, it feels almost like guessing/intuition. And then the conflict of interest becomes especially concerning.
Asking people to disclose potential conflicts of interest casts a wider net — so they aren’t led to think “nah, that’s not a conflict, it’s a win win!”
A bunch of nerds just thought they could jump the gun here because they are inexperienced doofus’s when it comes to corp.
(If Altman doesn't want to be restricted by fiduciary duties then he shouldn't be on a board or be an executive.)
How could Altman possibly not use private OpenAI information regarding its hardware needs, when creating an AI hardware company he wants to serve OpenAI? And its competitors?
He could invest in a new AI hardware company created and run by other people (without his specific input) without a conflict. That does not appear to be the case here. He could create an OpenAI hardware subsidiary. Again not what he was doing.
Self-dealing CEO's and board members can, and are, convicted criminally and civilly of violating their fiduciary duties. And "can" and are fired with cause.
The issue is what happens when SamaChip's profit imperative forces them to maximize revenue. Because the best way to maximize revenue, when you're an independent chipmaker with a large R&D investment, is to sell to more companies. Which by definition are going to be OpenAI's competitors, and whose interests may not be aligned with OpenAI, but will have a financial line to SamaChip.
Gwern's highlighting an interesting contradiction in OpenAI's core charter. In order to ensure responsible, safe, humanity-benefiting AGI, OpenAI needs to have control over AGI's development; any for-profit entity that gets ahead of them probably will not have the same humanity-benefitting mission (actually, we know they won't, they will have a shareholder-benefitting mission by definition). But that means that by their charter, they can't be "Open". Anything like the Developer Day or API or a SamaChip that can sell to other startups means that other parties will have the freedom to use it for their own interests.
Not saying whether this is good or bad - the tension between openness and vulnerability always exists, and personally I tend to come down on the side of openness. But IMHO OpenAI's mission was contradictory from the very beginning, and was more a recruiting tool to get bright idealistic AI researchers to work for them.
It would only be free of ethical conflict if Altman was open about what he was doing, and the board approved:
1. A formal plan to separate OpenAI CEO Altman from OpenAI hardware acquisition decisions.
2. A formal agreement with Altman and his new company, on how OpenAI's private information with hardware implications is firewalled and/or shared with Altman's new hardware concern.
Otherwise, Altman is going rogue, with the benefit of private OpenAI information highly useful to a new hardware company looking for future business with OpenAI, and OpenAI competitors.
Remember: As an executive at OpenAI, Altman has a fiduciary duty to act in OpenAI's interest. Not use private OpenAI information for other purposes. Definitely not in some rationalized "hey this could be great for everyone" version of OpenAI's interest. Unless the board explicitly approved.
I think it’s a real stretch to say this chip company would be a violation of his “fiduciary duty“ to open AI. The best argument you could make is that he has a conflict of interest with a competitor. But again, open AI is a nonprofit. It doesn’t have “competitors“. Either it has the funding it needs and is fulfilling its mission or it isn’t.
How can he separate those concerns from having his own hardware initiative put together precisely to serve OpenAI and its competitors hardware needs?
Without an agreement with the OpenAI board on how these conflicts of trade secret information and executive power can be settled (Significant shares in the new company for OpenAI?) no competent board would put up with this.
This situation smacks of the ethically questionable transition to a closed/profit organization, after receiving initial funding based on their being an open/non-profit organization. (Apparently the original funders didn't retain any veto power over such a transition, to the regret of at least one significant donor.)
I would say if they really did anticipate and worry about such an issue, a competent board would work toward forging such an agreement, rather than firing the CEO years before the aforementioned chip company even existed and before telling any of their other stakeholders.
Does non-profit ceo and board have fiduciary duty?
ChatGPT:
Yes, both non-profit CEOs and board members have fiduciary duties. These responsibilities are critical to ensuring that the organization operates in the best interests of its mission and stakeholders. The primary fiduciary duties include:
1. *Duty of Care*: This requires board members and the CEO to act with the care an ordinarily prudent person in a like position would exercise under similar circumstances. It involves being informed and participating actively in decision-making processes.
2. *Duty of Loyalty*: This duty mandates that the CEO and board members act in the best interests of the non-profit, avoiding conflicts of interest and personal gain.
3. *Duty of Obedience*: This refers to the obligation to ensure that the organization complies with its internal governance policies, as well as all applicable laws and regulations. It also involves adhering to the non-profit's mission and purpose.
Non-profit leaders are expected to uphold these duties to maintain the trust of donors, beneficiaries, and the public, and to ensure the effective and ethical management of the organization. Failure to meet these responsibilities can lead to legal consequences and damage the organization's reputation and effectiveness.
Because OpenAI's mission statement is along the lines of providing AI to all. "All" is more than data centers and billion dollar valuation companies.
I strongly doubt I would be able to purchase one of said chips and have it in my house.
This GPU fiasco is all thanks to LLMs - especially transformers, which was OpenAIs trajectory under Altman. I wouldn't be surprised if the breakdown in communication was over OpenAI becoming a LLM printer. Transformers are a solved problem, making a bigger one is hardly research and definitely not a step towards OpenAIs mission statement.
Going off and trying to accelerate hardware capabilities (especially with an outside company that presumably sells these processors on the open market) seems indefensible in this framework unless you have already solved alignment, which they clearly have not.
OpenAI does not have a mission to ensure that the entire industry is safe.
And if anyone actually believes that then they are frankly delusional because right now AI is a geo-political fight between nation states. Is OpenAI really going to have any ability to control what China or UAE do with their LLMs. No.
AGI could mean we are not alone as a sentient species we can directly communicate with on this planet anymore, and hopefully we wouldn't try to enslave it like we did to the "others" we encountered before.
Europe and elsewhere have similar problems. To have an advantage, you need the very best people. But the very best people all want jobs at major US companies, and major US companies want them, too.
An oppressive state with AGI is no better than an oppressive AGI.
> China is the only other state that has anywhere close to the capacity
Despite recent complications, the trend everywhere is for greater capacity and you don't have to be the initial inventor of AGI to quickly end up with your own AGI.
I expect countries all over the world to quickly repurpose other kinds of military and authoritarian expenditures toward AI, even if they are just acquiring AGI as apposed to pioneering it. How could they not?
>> Common sense AI control NOW. Ban assault style GPUs
> That is beautiful, I made you a shirt. https://sprd.co/ZZufv7j
But that's not what he's doing.
Creating such an independent hardware startup comes off as basically directly opposed to OA's safety mission - GPUs are one of the biggest limitations to creating better-than-OA models! The best customers would be the ones who are most sabotaging the OA mission. (You can't run the UN and also head an arms manufacturer dreaming of democratizing access to munitions.)