Vitalik Buterin (비탈릭 부테린), a co-founder of Ethereum, said AI still lags humans in its ability to judge documents.

Cryptopolitan, a blockchain media outlet, reported on Oct. 29 that Buterin wrote on X, formerly Twitter, that AI has vast knowledge but has limits in judging what a piece of writing is actually trying to say.

His point is more clearly seen in documents that have not been made public than in texts covering well-known material. If the topic is familiar, a model can grasp context based on years of accumulated data. But new essays or documents offer too little reference material. In such cases, the model must rely on the given sentences to understand the content.

The criticism also touches on practical work in the cryptocurrency industry. In areas where interpreting new documents is important, such as white papers, smart contract audit reports and governance proposals, relying only on AI summaries can cause subtle flaws to be missed. This “surface-level reading” is the failure point Buterin warned about.

Buterin has recently mentioned in succession how AI could affect cryptocurrency security. He warned that AI could weaken parts of cryptocurrencies’ core defense systems within 2 years and backed tools in which humans intervene in the final judgement. Proposals such as introducing “paranoia mode” to AI chats and his previously suggested “compute winter button” concept stem from the same concern.

Buterin did not take a negative view of AI overall. He argued in July that AI is not replacing all human abilities at once but advancing ahead of them one by one. In the same month, in an experiment to find the author of an anonymously written Ethereum document, an AI tool identified his writing style.

The remarks are interpreted as a warning that the scope and limits of using AI should be examined. They are especially significant for development teams and traders who rely on AI summaries without reading enough of the original text. Some investors, meanwhile, believe AI has already surpassed top human abilities in areas such as mathematics, and debate over AI’s capabilities and limits is expected to continue.

I find AI to be extremely knowledgeable, and so exceptionally able to detect references to almost anything, but not very wise, and so it only makes the most surface-level judgements about what point some writing is making. This is particularly true when it's analyzing writing that it just read for the first time, as opposed to analyzing writing that's already in its training data (along with lots of analysis of that writing by other people) AI's knowledgeableness is easier to see than its limited wisdom; it's perhaps even the very definition of wisdom that it cannot be easily verified (as opposed to intelligence).