Vitalik Buterin Explains Why LLMs May Not Be Enough to Match Humans

Vitalik Buterin, co-founder of Ethereum, argued on X today, July 20, 2026, that comparing humans and AI on a single scale like “how smart” they are, misses the point. Instead, he says both humans and machines are collections of many different abilities, mental, physical, creative, and social.

Historically speaking, machines have been able to outperform humans in some narrow areas, while humans kept a wide lead elsewhere. The big question that has been raised now is: will current AI trends (LLM plus the add-ons) eventually cover all these human strengths, or will there be more waves of innovation before machines truly match us?

Three Waves That Change Everything

According to the post on X, Vitalik framed technological progress as a series of breakthroughs that each expanded machines’ reach.

• The Industrial Wave: Machines learned to repeat the same physical tasks millions of times, allowing modern manufacturing.
• The Computing Wave: Calculators and computers automate formal, rule-based thinking which includes math, code and logic.
• The LLM Wave: Large language models began handling complex, example-driven tasks, writing, reasoning and even some physical tasks when paired with robotics or agents.

Even after these waves, Vitalik notes, people still do things that machines cannot. But the LLM wave blurs that line. Some argue that it is the real deal and will soon fill the remaining gaps. Others think more waves are needed to fill these gaps. Vitalik, however, sits in the middle. He is optimistic about the big gains but is unsure whether LLMs plus extensions will finish the job or not.

How AGI is Defined Here

Vitalik (and the analysis he relays) uses a practical definition of AGI. According to him, a system that, if uploaded into robot bodies after humans vanished, could independently continue civilization. This is stricter than “automating 90% of jobs.” It matters because it marks a point where AI is not merely a tool but it could be a long-term steward of human infrastructure and culture, a real point of no return.

Centaur Thinking: When Humans and Machines Team Up

Vitalik borrows the chess analogy: machines beat top humans decades ago, but human+machine teams (centaurs) remained superior for a long time. He imagines a similar twilight period for broader tasks: humans move up to high-level strategy while machines handle low-level details.

But the twist is deeper human-machine integration, brain interfaces, better interfaces that read subtle signals, and telepathic-like coordination, could erase the sharp divide between human and machine.

Vitalik sketches a future he likes, but warns it’s narrow. In that vision:

• Technology stays pluralistic: no single company or government rules everything.
• Humans can choose upgrades or stay as they are both paths that remain meaningful.
• Most of Earth becomes a protected, livable space; more adventurous projects go into space.
• Life gets easier (cheap food, better housing, cures), while humans still steer culture and politics.

Where Things Could Go Wrong

Vitalik lists several failure modes: a runaway AI that gains too much power, a single actor locking in decisive advantage, or human-machine integration that strips away what we value about humanity. He argues that these risks justify slowing down development, proposals that range from voluntary pauses to economic strategies that reduce the incentive to build ever-larger models.

Why “Open Weights” Matters

Vitalik favors open weights (public model weights) if they lead to slower frontier investment while enabling broader development of applications. He hoped openness will push compute-heavy breakthroughs to be less profitable, giving society time to set rules without centralizing power.

A Cautious, Practical Takeaway

Vitalik’s message is both helpful and alarmed. According to him, AI could lift everyone up, but that future is not guaranteed. He wants society to move carefully, preserve pluralism, regulate incentives, and invest in ways to keep humans meaningful participants as capabilities expand. In short, powerful AI tools are coming, we should steer them deliberately instead of being swept away.