OpenAI and Elon Musk

By /

Translated on March 5, 2024Published on March 5, 2024

Original text:OpenAI and Elon Musk

We have always been committed to realizing the mission of OpenAI.

author

OpenAI’s mission is to ensure that all mankind can benefit from artificial general intelligence (AGI), which not only means building AGI that are both safe and beneficial, but also means working to create widely distributed benefits. Now, we will share our understanding of how we achieve this mission, as well as some facts about our relationship with Elon. We intend to reject all of Elon’s claims.

We realized that building AGI required far more resources than we originally expected

Elon suggested that we should announce an initial $1 billion funding commitment to OpenAI. So far, the nonprofit organization has raised less than $45 million from Elon and more than $90 million from other donors.

When they founded OpenAI in late 2015, Greg and Sam initially planned to raise $100 million. Elon suggested in the email that we should increase the amount of funding to avoid sounding hopeless. He suggested that we should start with a $1 billion commitment and promise that he would take on anything that no one else could provide.1

We spent a lot of time thinking about how to implement AGI. In early 2017, we realized that building AGI would require a lot of computing power. We started calculating the computing power that AGI might require. We all understand that to achieve our mission, we need more capital, billions of dollars a year, far more than we, especially Elon, think we can raise through nonprofits.

Elon and we both believe that in order to access these resources, a for-profit entity is needed.

As we have discussed before, to further the mission, Musk wants us to merge with Tesla, or he wants to have full control. Musk left OpenAI, believing he needed a competitor that could rival Google/DeepMind, and he planned to do it himself. He said he would support us in finding our own path.

In late 2017, Musk and we decided that in order to advance the next step of the mission, we needed to create a for-profit entity. Musk wants to have a majority stake, control of the initial board of directors, and serve as CEO. Halfway through these discussions, he stopped funding. Reid Hoffman filled the funding gap to maintain wages and operating expenses.

We were unable to reach an agreement with Musk on a for-profit entity because we believed that giving any individual absolute control over OpenAI was contrary to our mission. He later proposed merging OpenAI into Tesla. In early February 2018, Musk forwarded us an email suggesting that OpenAI should “rely on Tesla as its source of profit” and commented that “exactly correct… Tesla is the only way to compete with Google.” Even so, there is little chance of becoming a force of opposition from Google. At least it’s not zero.” [2]

Musk quickly chose to leave OpenAI, believing that our chances of success were zero, and he planned to build a general artificial intelligence contender at Tesla. When he left in late February 2018, he told our team that he supported us in finding our own way to raise billions of dollars. In December 2018, Musk sent us an email saying,”Even if hundreds of millions were raised, it would not be enough. You need to invest billions every year immediately, or you’ll forget it.” [3]

Advance our mission by building widely available useful tools

We are making our technology widely available in a variety of ways that empower people and improve their daily lives, including our open source contributions.

We provide extensive access to today’s most powerful artificial intelligence, including free versions that are used by hundreds of millions of people every day. For example, Albania is using OpenAI tools to advance its accession to the European Union by up to 5.5 years;Digital Green is helping increase the income of farmers in Kenya and India, reducing the cost of agricultural extension services 100 times with OpenAI; Lifespan, Rhode Island’s largest medical institution, is using GPT-4 to reduce its surgical consent form from college reading to sixth-grade level; and Iceland is using GPT-4 to protect the Icelandic language.

Musk understands that achieving the mission does not mean open-source general artificial intelligence. As Ilya told Musk: “As we are closer to building artificial intelligence, it makes sense to reduce openness. In OpenAI, Open means that everyone should benefit from the results of already built artificial intelligence, but there is no need to share scientific knowledge…” To this, Musk replied: “Yes.”[4]

We feel very sorry about this because the person who happened to us was someone we deeply admired, who encouraged us to set our goals higher, then told us we would fail and start a competing company, and when we started making substantial progress on OpenAI’s mission without him, he filed a lawsuit against us.

We are currently focused on advancing our mission and there is still a long way to go. As our tools get better and better, we look forward to putting these systems into use so that everyone can benefit from them.

[1]

From: Elon Musk<

>
Attn: Greg Brockman<

>
CC: Sam Altman<

>
Date: Sunday, November 22, 2015, 7:48 pm
Theme: Follow-up after telephone communication

Blog sounds good, provided that we make some adjustments to make it more neutral and not focus too much on YC.

I suggest that we make blogs more attractive and attract public attention-there is a lot of value in getting the public to support us-and then we can make a longer, more detailed, more in-depth version for recruitment, and we can provide links at the end of the public-facing version.

We need a figure bigger than $1 million to avoid looking hopeless in the face of investment from Google or Facebook. I think we should say we start with a $1 billion commitment. This is true, and I will make up for anything that others fail to provide.

The template looks okay, except that it needs to be changed to the default equity cash bonus, which can be converted to YC or possibly SpaceX stock (we need to know how much this will be).

[2]

From: Elon Musk <

>
To: Ilya Sutskever <

>, Greg Brockman <

>
Date: Thu, Feb 1, 2018 at 3:52 AM
Subject: Forwarding: Today’s Top AI Institutions

Exactly. We may not want this, but between me and

Tesla seems to be the only way to compete with Google. Even so, the possibility of becoming Google’s check and balance is slim, just not zero.

Emails that started forwarding:

From:

<

>
Attn: Elon Musk<

>
Date: January 31, 2018 at 11:54:30 PM PST
Title: Re: Today’s Top Artificial Intelligence Institutions

Working at the forefront of artificial intelligence is unfortunately costly. example,

In addition to DeepMind, Google also has Google Brain, Research and Cloud. They also own Tensorflow, TPU, and account for about one-third of the research results (in fact, they host their own artificial intelligence conference).

I also strongly suspect that computing power will be a necessary condition (or even a sufficient condition) to achieve AGI (General Artificial Intelligence). If historical trends are any indication, advances in artificial intelligence are largely driven by systems-computing, data, infrastructure. The core algorithms we use today remain roughly the same as they were in the 1990s. Not only that, any algorithmic advancement published in a paper can be re-implemented and incorporated almost immediately. On the contrary, if there is no scale to expand the influence of the algorithm, the progress of the algorithm itself will be inert.

In my opinion, OpenAI is burning money right now, and its funding model cannot reach the scale to compete seriously with Google, an 800 billion company. If you can’t compete seriously but continue to study openly, you may actually be making things worse and helping them for free, because any progress will be fairly easy for them to replicate and apply immediately at a large scale.

The shift to for-profit may create a more sustainable source of revenue over time, and having the current team may lead to significant investment. However, building a product from scratch distracts the focus of artificial intelligence research and takes a long time. It’s unclear whether a company can “catch up” to Google’s size, and investors may put too much pressure in the wrong direction. The most promising option I can think of, as I mentioned before, is for OpenAI to use Tesla as its cash cow. I believe that with other large companies (such as Apple? Amazon？）Cooperation will fail due to company DNA incompatibility. To use the rocket analogy, Tesla has built the “first stage” of the rocket with the entire Model 3 supply chain, its on-board computers, and continuous network connectivity. The “second level” will be a fully autonomous driving solution based on large-scale neural network training, and OpenAI’s expertise can significantly help accelerate this process. With a fully autonomous driving solution that can run in 2-3 years, we can sell a lot of cars/trucks. If we do well and the transportation industry is big enough, we can increase Tesla’s market value to around $10 trillion and use this revenue to provide appropriate scale funding for artificial intelligence work.

I don’t see any other approach that has the potential to get sustainable Google-level funding within ten years.

[3]

From: Elon Musk<

>
Attn: Ilya Sutskever<

>, Greg Brockman <

>
CC: Sam Altman<

>,

<

>
Date: Wednesday, December 26, 2018 12:07 pm
Theme: I think I should emphasize again

My assessment of the probability that OpenAI will remain relevant to DeepMind/Google without making significant changes and adding resources is 0%. It’s not 1%. I wish this were not the case.

Even if hundreds of millions of funds could be raised, it would not be enough. This requires an immediate investment of billions of dollars every year, or forget it.

Unfortunately, the future of mankind is now in control

in the hands.

They did far more than that.

I sincerely hope I am wrong.

Elon

[4]

Repeat: Congratulations on the successful launch of Falcon 9

3 messages

From: Elon Musk<

>
Attn: Sam Altman<

>, Ilya Sutskever <

>, Greg Brockman <

>
Date: Saturday, January 2, 2016, 8:18 am
Subject: Retweeted: Congratulations to the Falcon 9

Start forwarding messages:

From:

<

>

Attn: Elon Musk<

>
Date: January 2, 2016, 10:12:32 AM CST
Theme: Congratulations to the Falcon 9

Hey, Elon

Here,

Happy New Year!

The successful landing of Falcon 9 is a remarkable achievement. Now is the time to expand the fleet!

I’ve noticed that you (along with Sam and others at OpenAI) have done a lot of interviews recently, praising the advantages of open source AI, but I think you should be very clear that this is not a magic solution that can easily solve AI security issues? There are many powerful arguments that the approach you are taking is actually very dangerous and may even increase global risks. There are some more obvious points set out in this blog post that I’m sure you’ve read, but there are other important considerations:http://slatestarcodex.com/2015/12/17/should-ai-be-open

I would love to hear your rebuttal to these views.

sweet

From: Ilya Sutskever<

>
Attn: Elon Musk<

>, Sam Altman <

>, Greg Brockman <

>
Date: January 2, 2016, Saturday 9:06 am
Subject: Reread: Congratulations to Falcon 9

The article focuses on a situation where AI breaks through quickly: If AI breaks through quickly and secure AI is more difficult to build than unsecured AI, then by opening up everything, we can let those with ill intentions and have access to large amounts of hardware resources. Easily build an unsecured AI, which will lead to rapid breakthroughs in AI.

As we get closer to building AI, it will become more reasonable to gradually reduce openness. Open in OpenAI means that everyone should benefit from the AI results that have been built, but there is no need to share the scientific research process (although in the short and possibly medium term, sharing everything is undoubtedly the most correct strategy for recruitment purposes).

From: Elon Musk<

>
To:

Ilya Sutskever <

>

Date: January 2, 2016, Saturday 9:11 am
Subject: Reread: Congratulations to Falcon 9

well

author

See all posts

Scroll to Top