Two details stand out. Each result included a proof written in Lean, a formal language that allows a computer to check mathematical proofs step by step. OpenAI estimated the computing cost for all ten results at roughly $2,000 at Sol API rates.
Some caution is warranted. Lean confirms that each proof is valid, but peer review must still assess whether each formal statement captures the original problem and how significant the result is. As of mid-October, I could find no public confirmation that any of the ten results had completed a formal journal peer review, and OpenAI did not say how many problems it attempted without success. Still, a proof a computer can check is harder to dismiss than an announcement.
The familiar comparison between AI and a calculator now feels incomplete. A calculator performs the operations we select. A system that produces a new proof does work that requires reasoning, exploration, and the recognition of which connections are useful.
Expertise Does Not Confer Immunity
We have usually imagined technological displacement as something that happens to routine work. A machine performs repetitive labor. Software automates bookkeeping. A computer replaces a long calculation. Under that view, people stay secure by moving toward more demanding intellectual work. The Astra results challenge that assumption. These systems do not need to match everything an accomplished mathematician or writer can do to change a profession. They only need to produce enough valuable work, well enough, to shift expectations, opportunities, and pay.
The estimate of $2,000 for ten results suggests how inexpensive some future mathematical discovery could be. Exceptional ability may remain exceptional even as it loses some of its scarcity.
Writers Face the Same Question
Writers have long relied on dictionaries, thesauruses, editors, and word processors. Those aids left the writer responsible for the substance.
AI can take on more of that substance. It can suggest an essay’s structure, develop a scene, propose dialogue, or revise an argument. Writers will have to consider that some work they regard as distinctly theirs can also be done by a machine.
It is tempting to settle the matter by declaring that AI will never write anything original or meaningful. That is a prediction and a weak foundation for a career.
Mathematics and fiction differ in one important way: how success is assessed. A formal mathematical proof can be checked mechanically against its stated assumptions. A novel has no comparable test for whether it succeeds with readers. Whether it works depends on craft, experience, taste, and what readers bring to it.
That difference cuts both ways. It may slow AI’s progress in fiction because a system has no clear signal indicating when a story has succeeded. It may also lower the standard AI has to meet. A proof must be correct. For some commercial fiction, sustaining readers’ interest may be enough to compete.
Progress in mathematics does not set a timetable for fiction. But writers have reason to pay attention when another demanding discipline confronts this change first.
Adaptation Is Uncomfortable
Telling people to adapt is easy. Experiencing it is harder.
A mathematician may have spent years learning to solve a particular class of problems. A writer may have spent decades developing a voice that a machine may eventually approximate. Both may feel that something important has been diminished, and that reaction is reasonable. However, that does not determine what the technology will do.
I have changed professions before. Before I wrote novels, I worked as a physician associate in vascular surgery and later as a technology executive at Teradata, where I worked on large-scale data systems. Each change meant accepting that the skills I had worked hard to acquire mattered less in the new setting than I had hoped. What carried over was judgment: knowing which question to ask and recognizing a wrong answer.
Humanity has repeatedly built tools that reduce the labor required for valued tasks. We have rarely agreed to preserve that labor because mastering it was difficult.
For mathematicians, adaptation may mean working with these systems: choosing problems, examining arguments, and understanding results well enough to build on them. For writers, it may mean directing a broader creative process, applying sharper editorial judgment, or making the work’s human circumstances part of its appeal. Some will continue working without AI. Others will use it heavily. Both will work in a world where this capability exists.
What We Don’t Know
It’s comforting to believe AI will always remain an assistant and that people will always keep the most important work. We do not know that.
We also do not know that greater machine capability renders human intellectual effort pointless. People do mathematics for understanding and satisfaction as well as for discovery. People write to express experience, examine ideas, and reach readers, as well as to earn a living. Those purposes can survive changes in how the professions operate and compensate their practitioners.
The practical approach is to learn what these systems can do and to test them directly. For a writer, that might mean drafting the same scene both with and without AI and comparing the results. We will also need standards for verification, attribution, and responsibility.
This summer’s announcement points to a future in which producing some research-level mathematical results could require surprisingly little additional computation. The full cost of discovery remains high, and the results still warrant scrutiny. But the possibility is consequential: work once dependent on scarce human expertise may become far more widely available. Mathematicians and writers will have to decide how to work in that world.