Axiom 將數學社群歷經二十年建立的 Stacks Project 轉成 Lean,把數百萬行內容交由電腦逐一檢查。
原作者指出,這項成果保留了各項數學結果彼此依賴的結構。
AI 伴隨問答
針對這篇報導有疑問?可直接點選快捷問題或向 AI 追問背景脈絡與技術細節:
Stacks Project 將數學成果與相互依賴的結構保留在 Lean 中,呈現 AI 助理如何在可由電腦檢查的基礎上延伸既有數學工作。
Everyone assumes AI-written code at scale means a giant pile of slop. Me too at times.
Axiom just put out the best counterexample I've seen.
They took the Stacks Project, the reference every algebraic geometer works from, built up by the math community over two decades, and turned the whole thing into Lean.
Millions of lines, and every single one is checked by a computer. It either holds up or it doesn't compile.
What gets me isn't even the size, it's the structure. Stacks is this massive web where every result leans on a bunch of others.
Axiom compares it to the glass pyramid at the Louvre, all these individual pieces that only make sense once you see how they fit together.
They kept all of that intact, just in a form a machine can check.
They ran their agents like a software team to do it, with some owning the architecture, some the core definitions, some the individual proofs, and humans at the top.
To be clear, the math is still human work.
People wrote Stacks.
What's new is that it's now a verified foundation AI agents can actually build new math on.
Link is below ↩️