Dialogue with Vitalik: After donating hundreds of millions of US dollars, how to start reflecting on EA?

How to help more people with limited funds? Effective Altruism (abbreviated as "EA") starts from this simple question, pushing people to compare the effects of different charity projects, ranging from donating mosquito nets to prevent malaria, to later investing hundreds of millions of dollars in AI safety research.

The collapse of SBF and FTX once pushed this movement into a credibility crisis. Recently, the resignation of AI researchers and controversies triggered by safety assessment institutions and donation networks have drawn public attention to EA again. When some people regard preventing AI from destroying humanity as the most worthy cause to invest in, how do their judgments affect research directions and capital allocation? What interests and power issues are involved in this process?

In the first part of this episode, through the perspective of tech journalist Afra, we learned how EA evolved from a movement discussing "how to do good deeds more effectively" to gradually influencing today's AI research, capital flows, and even political agendas. The reason why these ideas can exert such influence today is inseparable from the support of huge amounts of funds behind them.

Therefore, in the second part of the program, Silicon Valley 101 had a conversation with one of the important donors — Vitalik Buterin, co-founder of Ethereum. In 2021, he donated a batch of cryptocurrencies to a number of charitable institutions for fields such as pandemic relief, global health and scientific research. According to his recollection, the part donated to the Future of Life Institute (FLI) was cashed out for about 400 to 600 million US dollars. FLI also clearly stated that Vitalik is the largest donor of the institution so far. However, FLI's subsequent focus changed: from researching all fields that fight against existential risks to humanity, it shifted to fully betting on the single direction of AI safety, and increasingly moved towards policy actions.

Vitalik told us in the program that he tried to communicate with FLI, hoping that the other party could expand its areas of concern, but FLI refused, which became the starting point of their divergence. In his view, EA's ideas are very effective in the field of man v.s nature, but when stepping into the field of man v.s man, the situation will become much more complex.

Vitalik believes that there is still great uncertainty about whether AI has the risk of destroying humanity; people with different positions have different doubts about Anthropic's safety philosophy, and even within EA, there are two groups of people who support and oppose Anthropic; facing the progress of technology and various unknown risks, he has summed up a set of d/acc (Defensive Accelerationism) ideas, advocating priority to develop technologies that can protect people and decentralize power, such as cybersecurity, vaccines and air filtration, to improve society's ability to resist risks.

The following is a selection of the content of this conversation:

01

Vitalik talks about AI risks:

There are still too many uncertain points about whether it is valid

Yiwen: What was the reason for your initial exposure to EA? What attracted you at that time?

Vitalik: The first time I heard the two concepts of rationalism and EA was probably between 2013 and 2015, when both were in a very early stage. Eliezer Yudkowsky was on the rationalism side, he had a blog called LessWrong, which published a lot of articles called The Sequences, where they talked about all interesting topics including physics, morality and so on.

The group of effective altruists said: In fact, most of the charity we do has very good goals and very good intentions, but most of them are very low in efficiency. At that time, they discussed a lot of topics about health in Africa. They had a very famous statistic: if you use 2000 US dollars (at that time) to buy mosquito nets for people in very poor African countries, you can protect many people from malaria. Every 2000 US dollars you spend can save a person's life.

If that's really the case, then I have to spend my own money in such places too. Between 2016 and 2017, the price of Ethereum rose 10 times, so I sold a small part of my Ethereum and donated it to those organizations.

Insecticide-treated mosquito nets are one of the most effective interventions for global malaria control, and are also a symbol of EA's "mosquito net period". Image source: President's Malaria Initiative (public domain)

Yiwen: When you started to pay attention to EA, what was your thinking about AI safety? At that time, you donated money and contacted some organizations, what was their general involvement in AI safety?

Vitalik: At that time, my thoughts on AI safety were still very abstract. I knew that AI was theoretically a very important topic, what they said might be true, and AI safety might determine the future destiny of mankind. But no matter whether this possibility is 100% or 0.1%, at that time, there were very, very few people thinking and studying these topics.

In 2015, the AI we are talking about now hardly existed. The so-called AI we worried about in 2018 might be cameras and digital surveillance with AI assistance. Between 2020 and 2023, the release of GPT-3 changed a lot, and AI became a real topic. After ChatGPT came out, we began to understand: AI may really develop very, very fast, and AI will really bring us a lot of changes.

Before 2022, the only answer was that the more we can help research and the more humans know, the better for us. But after 2022, it changed from an abstract topic to a concrete one.

Hongjun: Will you worry about AI safety issues?

Vitalik: People ask me about P(doom) (the probability of human extinction) — the probability that AI or other risks will completely destroy our world and make human beings cease to exist — I will say 0.1. Why do I say 0.1? Why not 0.03? Why not 0.3? In fact, it is very difficult to be more specific, because I have great uncertainty from the very beginning, and there are many, many uncertain points.

First, how fast will AI develop? It may take 10 years to reach the so-called AGI that everyone talks about, or it may take 300 years.

Second, is the alignment problem that Yudkowsky talks about really a real problem? Will this problem solve itself easily? For example: At that time, we make a very smart AI, and we tell it that we hope to have a world without cancer, and this AI thinks the easiest way is to destroy the whole world and kill everyone. In fact, if you talk to Claude, GPT and modern AI now, you will know that current AI has a better understanding of context, they understand our meaning — this is a change that was difficult to predict 10 years ago.

But this does not necessarily mean that the alignment problem is not a very big problem. If you ask Yudkowsky now: Look at the current AI, if you tell them to cure cancer, they will not completely destroy the world, so is the alignment problem solved? He will say: No. This problem is only a small part of the alignment problem, there are many other parts, and there are many risks that we have not solved at all and do not know at all. There is also a lot of uncertainty here.

Third, there is political uncertainty. Because now different countries, different people and different companies are all working on AI. In a competitive situation, everyone has different goals, everyone hopes for safety at the same time, and everyone also hopes that they can win. If, for example, when AI becomes very dangerous, World War III breaks out at the same time, the most dangerous thing may not be due to any technical problem, but because at that time, everyone will think the least about safety, and only think that "winning is the most important thing for me".

So there is technical uncertainty multiplied by another technical uncertainty, multiplied by political uncertainty, and multiplied by many more. It is very difficult to predict whether AI will really be dangerous to us.

02

Vitalik's donations:

Dogecoin, pandemic and divergence

Hongjun: I noticed that you donated to many EA organizations at the same time on May 12, 2021; in 2017, you donated to organizations such as the Machine Intelligence Research Institute (MIRI) founded by Yudkowsky at that time. Could you talk about what you were thinking when you donated to many organizations on that day in 2021? How did you make this donation decision?

Vitalik: This story is very interesting. In 2020, Dogecoin was very successful, and everyone started to issue their own Dogecoin, the largest of which was Shiba Inu (SHIB).

The official logo of Shiba Inu SHIB was born in August 2020, with a circulation of 1000 trillion. In early May 2021, the sharp rise in the price of Shiba Inu coin attracted market attention. Image source: shib.io

They also did a very interesting thing: they put half of the supply of their coins in my wallet. At that time, it was during the pandemic, so I didn't do anything for a long time. But one day I found that the market value of these coins really reached 8 billion US dollars, which was really very, very exaggerated. So I chose: Well, their market value may become so high in a week, but maybe the bubble will burst in a week, and they will become very low again.

I did a lot of complicated work to take out those Dogecoins. At first I quickly sold them for Ethereum, and I probably earned 70 million US dollars worth of Ethereum. I donated most of it to the Against Malaria Foundation, the organization that buys mosquito nets. This money is probably the maximum amount of money I can get, so it should be given to a relatively safe and uncontroversial organization. I also gave some to other longevity health organizations.

Because my market depth was not enough, the liquidity was not enough for me to sell so many Dogecoins, so after I sold them for Ethereum, I still left a lot of Dogecoins. So I decided: Well, maybe these coins will fall very fast and very badly now, I will donate the remaining coins to some other organizations that I think are good, and it will be fine if they can get a little when they sell them by themselves. In my mind, they could get at most more than 20 million US dollars.

So I gave half to CryptoCOVID Relief, an Indian organization fighting COVID, and the other half to FLI (Future of Life Institute). At that time, FLI was an organization that studied and began to fight against all existential risks that could end human beings. They studied every major risk, such as biology-related, AI-related, and nuclear war-related risks. They introduced me to one of their plans, telling me that in theory they could reach up to 100 million US dollars, but 20 million would also allow them to do a lot of things. So I said, okay, I'll give you my Shiba Inu coins.

Three months later, I found that those Dogecoins did not fall. The first organization, CryptoCOVID Relief, sold all these coins and got 470 million US dollars. I don't know exactly how much FLI got, maybe 400 million, maybe 600 million.

It completely exceeded my imagination. If I could have anticipated such a result, I would have done things very, very differently. So the next year I chatted with Sandeep from the CryptoCOVID organization, convinced him to return half of his money to me, and then we started doing our own anti-pandemic work, including researching filters, ultraviolet light, masks, and long COVID research. We donated to some open-source vaccine companies, one is RADVAC (open-source nasal spray vaccine project), and the other is PopVax (Indian biotechnology company).

Around the beginning of 2022, everyone thought the COVID pandemic was over, so we started to fight against all airborne diseases. We used 200 million US dollars to do a lot of things. Sandeep used his 270 million US dollars, which can be said to have upgraded India's entire medical manufacturing capacity.

Around this time, after GPT-3 came out, FLI's thinking changed a lot. First, they changed from caring about all existential risks to only caring about AI; second, they changed from a research attitude to a political action attitude. They began to use a very specific strategy, thinking about which laws to support and which politicians to support. They started doing a lot of things that I either disagree with, or that I would not spend my own money to do.

I communicated with them two or three times, I told them that my original goal was to support all fields against existential risks, and I hoped they could expand their scope. But they refused, they wanted to put all their money into an AI-centered and policy-centered direction. At the same time, those laws they called for also came out, many investors from a16z and Silicon Valley were very unhappy, and they began to regard AI safety as their main enemy, which probably started in 2023.

03

From saving lives to influencing policies

Are EA's ideas still effective?

Hongjun: Do you think your ideas have changed now? How to donate and how to donate properly? In fact, after the donation, we also see that the whole world is changing — AI is growing rapidly, and the direction they invest this money in is not quite the same as the original intention of going to Africa to cure diseases at the beginning.

Vitalik: I think there is an important concept in EA, that is neglectedness, which means: are there other people already doing this? About 10 years ago, I thought that every field related to EA and rationalism was very neglected at that time.

But that's not the case now. Now AI safety is a very high-level topic, as is the topic of longevity health. However, global public health actually suffered a lot in 2025 — because of what Elon Musk did and a political change in the United States, they previously supported many public health-related projects, which were quickly stopped, creating more problems. So for global public health, I think it has become more important now, and it is a topic worthy of support.

But most other fields are not neglected. It's not that there are too few resources invested, but that the processing methods are wrong, and the required solutions are completely different. Then solving these problems is not a research problem at all, it really starts to become a policy problem. But the biggest challenge of policy problems is that it is very difficult for you to know whether a policy is good or bad. So policy problems are very different from the problems that EA people thought about before.

Because the problems that EA usually thought about before are: if you donate to agency A, you may save 10 lives; if you donate to agency B, you will save 100 lives. Your challenge is to know which one is 10 and which one is 100. But now it is completely different, because you are supporting a policy, which may have good effects or bad effects, and in fact no one knows which effect is good and which effect is bad, which is very complicated.

This policy has some direct effects, that is, how it directly affects the incentives of those companies; it also has some indirect effects. For example, if a rich person publicly spends 500 million US dollars to support some political topics, other people may be very unhappy, they will think "a local tyr