Vitalik Buterin

Vitalik Buterin@vitalikbuterin · 4 天前赛道影响

A new type of "high-level programming language" that seems really worth trying to make, is a language that gets compiled to Lean (or HOL, or...) that is specifically about making it as friendly as possible for a human to read definitions and theorems. Not the proofs - as all that matters with proofs is that the proofs are correct - just the definitions and theorems. The intended use case is that AI outputs a blob of proofs, and you're trying to make it as easy as possible for anyone reading the output to understand what the actual precise claims are that have been proven

AI：偏主题投资表达，释放对基础设施与新叙事的偏多信号。

Vitalik Buterin@vitalikbuterin · 5 天前观点输出

I'm pretty sure my arguments did not depend anywhere on intelligence being everything - in fact the opposite, the issue is about how capabilities are very multi-dimensional, but at the same time "multi-dimensional" does not mean "we're definitely guaranteed to hold on to some special snowflake capability that nothing else will be able to match".

AI：延续个人公开立场，强调长期主义、执行效率与行业方向判断。

Vitalik Buterin@vitalikbuterin · 5 天前政策影响

But here we get to the thing that will confound looking at human and machine as separate units: the possibility of deeper and deeper integration between human and machine. This is the strand of technology that includes: computers being able to work much more symbiotically with us, picking up on all kinds of conscious and sub-conscious signals we release as an interface, brain-computer interfaces, computer-assisted telepathy that helps multiple people better understand each other and work together. The ultimate end is erasing the binary divide between human and machine and opening up any position on that spectrum. On one end, you sitting at a laptop terminal; on the other end, mind uploading, and also everything in between. The hope is that deeply integrated human + machine will outperform separated human + machine, and that this will remain the case all the way up to the technological ceiling. This is the best "far-future AI outcome that I actually like" that I can come up with: * We preserve global political and economic pluralism (no single company or government dominates) all the way up to the technological ceiling. * We humans each get the ability to either take technological upgrades to ourselves, and explore the universe at a level competitive with "pure" machines, or stay as we are, and get some ability to act in the world through having a bot listen to our instructions. * Most of Earth will probably become a highly-regulated nature preserve where anyone can continue living the kind of life they have now (without the pains and annoyances; most challenge will be healthy "good challenge"), and if you want to do more crazy things, you go to space. Make no mistake - this is a ride through a narrow corridor. Things could go off the rails in one of several different ways: AI-on-its-own could get too powerful relative to the alternatives; one country or one company could gain decisive advantage and decide to lock it in permanently, powerful technological ways of destroying each other could cause incalculable harm, or the human + machine integration approach itself ends up preserving not enough of our humanity, and so it ends up being "just another way to make powerful bots". For these reasons, I wish that we can ride through the narrow corridor at 60 kph and not 200 kph, which is why I feel warmly toward "slowdown" / "pause" proposals (including "give us a chance to hit the brakes" ideas like mutually assured compute destruction (as in AI 2040, also previously Dan Hendrycks, and others)). My ideal world is actually one where (i) open weights wins, and (ii) Dean Ball turns out to be completely right that open weights is decelerationist because it deters capex in frontier models [though it will enable more capex in peripherals and applications], because then we get slowdown without any of the downsides of doing it through heavier-handed political means. But also I don't think we should count on the world to naturally give us "convenient" laws of economics and physics. We do have some degree of agency over the laws of politics, plenty of ways to change incentives without putting a king in charge, and we should exercise that.

AI：偏政策推进，强调合规落地与规则明晰，对监管主线更敏感。

Vitalik Buterin@vitalikbuterin · 5 天前观点输出

But AGI is not the same as ASI (artificial superintelligence). To see the difference, let us zoom into what things might look like one year before the singularity. We also need to add a third line to the chart: the centaur (human + machine). An analogy here is chess. Machine surpassed the average human in the late 1970s, and surpassed the top human (Kasparov) in 1997. But human + machine continued to beat machine alone all the way until around 2017. What did the twilight of human + machine in chess look like? Actually, it did look like humans being relegated to more and more "high-level strategic thinking" functions, while bots managed the move-by-move. Very similar to vibe coding in 2026. Before we get to a point where there are no areas where our input is economically helpful, we will be in a regime where there are only a few such areas. And because we humans are scarce, those few areas will become the global economy's bottleneck - they will be the only thing you can't optimize down to near-zero cost (and latency), and so all our attention will go toward those tasks.

AI：延续个人公开立场，强调长期主义、执行效率与行业方向判断。

Vitalik Buterin@vitalikbuterin · 5 天前赛道影响

Then, we had a series of three eras of advancement, in which machines rapidly gained areas of capabilities that far exceeded our own: * In the industrial revolution, mass manufacturing, or generally the ability to perform the exact same action millions of times * With calculators and computers, the ability to do mental tasks that can be defined via an analytic-philosophy-style logical definition (eg. computer code) * With LLMs, the ability to do mental tasks (and physical tasks, eg. already self-driving cars) that can be defined via a large set of examples But even after all three waves, at least today, there is still a large range of capabilities that remain unique to humans. A big question facing us is: is the LLM wave "the real deal", in the sense that, as they keep getting better, they will eventually cover either all missing capabilities, or at least enough capabilities that the rest can be worked around, or are LLMs simply a continuation of the same pattern, and it will take some fourth or fifth wave to go the rest of the way?

AI：偏主题投资表达，释放对基础设施与新叙事的偏多信号。

Vitalik Buterin@vitalikbuterin · 5 天前赛道影响

Thread: how I think about the growth of AI capabilities and what it means for us as they keep getting stronger. I think the right model for comparing humans and AI is not any single dimension like intelligence or aggregate performance on economic tasks. Rather, I view both humans and machines as having a whole range of capabilities. This includes both physical (eg. doing things with our hands, walking) and mental (eg. strategic thinking, mental math, emotional intelligence). For most of human history, machines have only exceeded us in very few areas, eg. a strong early example is water and windmills. So we can look at the situation in 1500 like this:

AI：偏主题投资表达，释放对基础设施与新叙事的偏多信号。

Vitalik Buterin@vitalikbuterin · 6 天前观点输出

btw it's stage 2 https://t.co/U5UQkunZD9

AI：延续个人公开立场，强调长期主义、执行效率与行业方向判断。

Vitalik Buterin@vitalikbuterin · 6 天前观点输出

Vibe-coded a toy demo version of the "anon billboard with moderation" idea from https://t.co/fheLOlfF3F on aztec https://t.co/iIkwH3UbO4 It's early days but you can already do very interesting nontrivial stuff! https://t.co/NbfrBJ4FtU

AI：延续个人公开立场，强调长期主义、执行效率与行业方向判断。

Vitalik Buterin@vitalikbuterin · 7 天前赛道影响

RT @kassandraETH: spending my saturday testing... tornado AND railgun AND privacy pools deposits / withdrawals on the ethereum mainnet al…

AI：偏主题投资表达，释放对基础设施与新叙事的偏多信号。

Vitalik Buterin@vitalikbuterin · 10 天前赛道影响

RT @EFDevcon: Mumbai is calling. 🇮🇳 Global tickets for Devcon 8 are now live. This is the one to be in the room for. November 3-6 https:…

AI：偏主题投资表达，释放对基础设施与新叙事的偏多信号。