AI Is Centralized Power, Crypto Is Its Antidote
Vitalik says AI's power comes from compute and data, and the best AI is the biggest AI, which naturally tends toward centralization; Crypto's only role is to solve trust problems between people, so AI participates in Crypto's game, not the other way around.
The video won't play here. Listen to the audio instead:
The argument · tap a timestamp to hear it
AI can't help him with complex cryptography
Vitalik says he made an interesting discovery using AI himself: for doing things many people have done before but he hasn't, ChatGPT is very useful; but for doing truly cutting-edge things, like particularly complex cryptography, which only he and a thousand other people have done, AI is of no help at all. This showed him a two-sidedness of AI—it can let some people do things they previously had no idea how to do, which is relatively egalitarian; but on the other hand, ChatGPT itself is a very centralized thing, and when you use it you need to completely trust that it won't expose your data.
— Vitalik ButerinOpenAI's two-step compromise: first abandon open source, then abandon safety
Vitalik breaks OpenAI's story into two steps: first, for AI safety reasons, they said they would no longer open source, redefining "open" as the service being open, turning into close AI; second, this year they sacrificed safety for profit—after last year's conflict between the company and the board, Sam Altman won, and recently it was announced that it would go from nonprofit to for-profit, with the board's power reduced to just an advisor. He quotes an American founding father: you cannot sacrifice freedom for safety; if you do, you will find you have neither safety nor freedom.
— Vitalik ButerinLLMs are like biological life, not ordinary programs
Vitalik says there is a difference between LLMs and ordinary programs: in an ordinary program, if you randomly delete a line of code, the whole thing might stop working, it's very fragile; but with an LLM, making small changes doesn't affect its basic function, and the output only changes a little, so LLMs are very much like biological life. He also mentions Anthropic's research, looking at what concept each parameter represents in a relatively small LLM, and finding that you can specifically say this parameter represents red, that one represents plants, that one represents capitalism, that one represents Ethereum.
— Vitalik ButerinAI participates in Crypto's game, not the other way around
Vitalik says Crypto and AI have an intersection that has existed almost from the beginning but almost no one mentions: decentralized exchanges have many bots doing arbitrage and market making. He abstracts this example into a category—Crypto's role is to make a game, using smart contracts to execute the game's rules relatively safely; AI's role is to participate in this game, because if only humans participate, efficiency won't be high enough. He gives the prediction market Polymarket as an example: four years ago Augur's result quality was not high, this year it's much higher, because liquidity went from maybe 1 million, 10 million USD to 100 million, 200 million, and AI's reaction speed is much faster than humans.
— Vitalik ButerinAI's power comes from compute and data, naturally centralized
Vitalik says AI's power comes from two places: first, computing power; second, data. If you have more computing power, your AI will be stronger; if you have more data, your AI will also be stronger. So AI is a particularly powerful thing, but the best AI is the biggest AI, and if you want to make an AI, a very simple method is to put all the computing power and all the data together, in one place. Crypto is the opposite: communities and projects are very dispersed, and the Ethereum Foundation may have only 25% of its people in the US, with the rest everywhere else.
— Vitalik ButerinThe only way for humans to participate in superintelligence is to become part of it
Vitalik says something smarter than humans will definitely appear, maybe in ten years, a hundred years, a thousand years. There are two choices: one is that it becomes an independent thing that surpasses us; the other is that we can become part of this thing, with biology and machines merging together. He finds the second direction interesting, but it must be done the right way—if a technology like Neuralink is centralized and not open source, only sending all your information to a server, that means a big company could read your thoughts, which is very dangerous. He hopes for open-source software and hardware that respects safety.
— Vitalik ButerinIf smart people all stop doing Crypto, the industry has failed
Vitalik says AI's boom has indeed brought a certain cooling to Crypto. He divides the people who left into three types: the first type's goal is to make money or participate in the most important things for humanity; the second type cares about specific issues like currency and privacy for concrete reasons; the third type wants to build but does things with lower quality. He has a worry: if all the particularly intelligent people stop doing Crypto, then the people left in this field won't have any interesting ideas, and the only applications will be issuing a coin, making a small exchange, then issuing another coin, and this coin has an open dog. He says those things are fun, but if they are the only things, our industry can be said to have failed.
— Vitalik ButerinBefore this year there wasn't enough technology for ordinary people to participate
Answering "why doesn't the Web3 ecosystem have real applications," Vitalik says that before this year there simply wasn't enough technology. Some very important technologies weren't mature yet, security problems weren't solved, many privacy problems weren't solved. The only successful application was DeFi, and the reason is that DeFi can make a lot of money—if you have a 30% chance your account gets stolen, but winning increases your money tenfold, you might still participate. But if your goal isn't to increase your money tenfold, but includes your identity and participating in other applications, then with the technical difficulty and security problems unsolved, you won't participate. This year some problems are finally starting to be solved, so it's the best time to build meaningful applications.
— Vitalik ButerinIn their own words · checked verbatim
I do some particularly cutting-edge things, like some particularly complex cryptography, where it's possible that in some particularly complex cryptography fields only I and a thousand other people have done it, and in these fields I find every time that some AI is of no help to me at all.
我做一些特别前年的东西 比如一些特别复杂的密码学 就是有可能在一些特别复杂的密码学的领域 只有我和一千个其他的人做过 那在这些领域 我每次都发现 一些AI对我一点帮助都没有
Vitalik Buterin10:26
Then this year there's another problem, which is first they sacrificed their open source for safety, and this year they sacrificed their safety for their interests.
那今年有另外一个问题 就是第一他们是为了安全 牺牲了他们的开源 那今年他们为了他们的利益 就牺牲了他们的安全
Vitalik Buterin14:32
You cannot sacrifice freedom for safety; if you do, you will find you have neither safety nor freedom.
不能为了安全牺牲自由 如果你这么做你会发现你安全和自由两个都没有
Vitalik Buterin18:35
Crypto's role is that Crypto is a relatively safe base layer, through Crypto you can make a game, we can guarantee the game's rules are fair. AI's role is to participate. So it's not Crypto participating in AI's game, it's AI participating in Crypto's game.
Crypto的作用就是 Crypto是一个比较安全的地层 通过Crypto可以做一个游戏 我们可以保证这个游戏的规则 都是公平的 AI的作用就是参与 所以不是Crypto参与AI的游戏 是AI参与Crypto的游戏
Vitalik Buterin27:49
AI is a particularly powerful thing, but the best AI is the biggest AI. If you want to make an AI, a very simple method is to put all the computing power and all the data together, in one place.
AI是一个特别有权利的一个东西 但是你是最好的AI就是最大的AI 如果你想做一个AI也很简单的方法 就是把这个整个计算的力量和整个的数据都放在一起 放在一个地方
Vitalik Buterin28:51
Then maybe the only applications will be some of the things everyone has done for many years, which is that the crypto field will all become one situation: issuing a coin, making a small exchange, issuing another coin, making a small exchange, issuing another coin, and this coin has an open dog. That's fine.
那有可能唯一的应用就是一些 大家已经很多年做过的一些经营的 就是这个crypto一个领域 都会变成一个情况 就是做一些发个币 做个小利索 再发个币 做个小利索 再发个币 这个币有一个开的狗狗 那行的
Vitalik Buterin43:09
If there are some, for example, you have a 30% chance that you will hand over your account, or your account will be stolen, then if you win your money will increase several times. If you run into a problem, your money is all gone, then maybe you will still participate.
如果有一些比如 你有30%的概率 你会交自己的账户 或者你的账户会被投 那如果你赢了 你的钱会增加设备 如果你遇到了一个问题 你的钱都没了 那可能你还是会参与
Vitalik Buterin45:11
Figures
| Vitalik's local GPU | NVIDIA 4070 | 12:30 |
| ChatGPT paid tier monthly fee | 20 USD | 13:32 |
| Polymarket's liquidity this year | 100 million to 200 million USD | 26:47 |
| Share of Ethereum Foundation people in the US | 25% | 29:52 |
| Example probability of DeFi participant's account being stolen | 30% | 45:11 |
Glossary
- Layer 2
- Ethereum's scaling solution, processing transactions off the main chain to reduce fees.
- Farcaster
- A decentralized social network protocol, with the first layer being the decentralized network and the second layer being clients.
- Warpcast
- The most commonly used centralized client app in the Farcaster ecosystem.
- Polymarket
- A decentralized prediction market whose liquidity has grown substantially this year.
- Worldcoin
- A digital identity project involving Sam Altman, attempting to solve the problem of "who is human and who is machine."
How to listen
Founders and investors focused on the intersection of AI and Crypto, especially those who want to know how decentralized technology can combine with AI, and how Vitalik himself judges OpenAI.
The last 10 minutes about life, cats, marriage, and the US election, which is low in information density.