TL;DR
Get the little things that make your day delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
A November 2026 blog post by computer scientist Murat Buffalo describes writing as a way to externalize thought, test understanding and develop ideas. He argues that AI-generated prose cannot replace the thinking involved in writing for oneself, while acknowledging that this is an argument rather than a demonstrated result about all AI systems.
Computer scientist Murat Buffalo has published a blog post arguing that writing helps people think by putting ideas outside the mind, where they can be examined, connected and revised. The November 2026 essay also contends that relying on large language models to generate prose can obscure whether a person has done the reasoning themselves, though the post presents this as the author’s view rather than a finding from a reported study.
Buffalo traces his interest in writing to middle school, when he struggled to retain long passages. He says he began making extensive notes using Tony Buzan-style mind maps, a technique that had reached Turkey in the late 1980s. The notes helped him work out how pieces of information related, he writes; their value was not simply storing material but processing it.
He later connected that experience to advice attributed to computer scientist Manuel Blum. In the passage Buffalo quotes, Blum compares finite automata with Turing machines: both are theoretical computing models, but the latter can use a paper-and-pencil-like memory store. Buffalo treats the comparison as a metaphor for how writing can expand the amount and complexity of thinking a person can manage. It is not a formal claim that people literally change computational complexity classes when they write.
The post describes writing as a form of external memory, similar to “out-of-core” computing, in which a system uses storage outside its immediate working memory. Buffalo says that drafting a review of a research paper can expose gaps in his understanding and help him place the work in a wider context. He also recounts starting a 2018 blog post about similarities between Paxos and Nakamoto consensus without beginning with that thesis; he says the connections became clearer as he wrote and the post later helped seed a research paper.
Writing as a Check on Understanding
The argument matters to students, researchers and workers who use writing to organize complex material. A draft can serve not only as a record of a finished idea but as a way to find out whether the idea is coherent. Buffalo’s example of writing a paper review illustrates the distinction: feeling that a paper is clear while reading it is not the same as being able to explain its claims and place them in context.
The essay also addresses a practical question about AI-assisted writing. Buffalo argues that a language model may supply established knowledge and familiar framings but cannot be assumed to recognize a user’s private, incomplete intuition or make a novel connection on the user’s behalf. That is his assessment of current systems, not proof that models cannot contribute to creative work. The broader point is that polished output alone does not show that the person using it has understood or evaluated the underlying ideas.
For readers, the distinction affects how AI tools might be used in learning and research. If a person asks a model to produce a finished explanation before forming their own account, the result may conceal what they do not understand. Writing an initial draft first, then using AI to question or improve it, may preserve more of the writer’s own reasoning. The post does not test that approach against alternatives, but it makes a clear case for treating drafting as part of thinking rather than as presentation alone.
Popular right now
Dreame G10 Pro Wet-Dry Vacuum Mop
Apple AirPods Pro 3 Wireless Earbuds with Active Noise Cancellation
De'Longhi Classic Signature Espresso Machine with Adjustable Milk Frother
As an affiliate, we earn on qualifying purchases.
From Mind Maps to AI Drafts
Buffalo’s essay moves from a personal account of note-taking to examples from academic work and then to the growing use of language models. Its central concept is externalization: a person records thoughts in a form they can inspect, rearrange and return to later. In the author’s account, this can support both immediate reasoning and longer-term recall.
He says he often revisits his own blog posts because writing them required effort, and that rereading helps him reconstruct the ideas. He also quotes Leslie Lamport’s line, “If you think without writing, you only think you’re thinking.” The remark reinforces the post’s thesis, but it is a maxim, not a research result established by the essay.
Buffalo contrasts this process with producing a document through an AI model. He points to Amazon’s six-page narrative memo as an example of writing associated with careful reasoning, and invokes Jeff Bezos’s statement that writing such a memo requires clear thinking. Buffalo argues that generative AI weakens the inference: a long, well-formed document can now be produced without demonstrating that its user did the thinking. The post offers no new reporting on Amazon’s internal practices or measurements of AI’s effects.
““Without writing, you are reduced to a finite automaton. With writing you have the extraordinary power of a Turing machine.””
— Murat Buffalo, quoting Manuel Blum
What the Essay Does Not Establish
The post is a personal and argumentative essay, not a controlled study. It does not quantify how much writing improves comprehension or memory, compare handwritten notes with digital tools, or establish that its examples apply equally across people and tasks. Its computational analogy is explanatory, not a literal measurement of human thinking.
Buffalo’s claims about AI are also not tested in the post. It does not define which models or tasks he has in mind, or provide evidence that language models cannot make new connections. The essay’s narrower warning is that a model’s fluent output should not be mistaken for proof that a user has reasoned independently.
A Draft Before the Model
Buffalo closes with practical advice: begin by producing a rough draft without stopping to correct every error or make the prose presentable. He calls this an initial “zeroth draft” that gives the writer material to react to. In his account, even an imperfect writing session can help because it makes problems visible and creates a basis for revision.
For research work, he recommends keeping a working notes file as a lab notebook, recording questions, tangents and developing thoughts while the work is underway. The immediate next step for readers is not an announced study or policy change; the post offers a method to try. Whether drafting before turning to AI leads to better understanding for different users remains an open question.
Key Questions
What does Buffalo mean by writing as a thinking tool?
He means that writing moves ideas into an external form that a person can inspect, organize and revise. In his view, that process can reveal gaps in understanding and help ideas develop.
Does the essay prove that writing improves thinking?
No. It presents Buffalo’s experiences, examples and interpretation, but reports no controlled study or measured effects. Its claims should be read as an argument rather than experimental proof.
Is Buffalo saying people should not use AI for writing?
No categorical ban is proposed. He warns that asking a model to produce prose can replace a person’s own initial reasoning, and argues that writers should not mistake fluent output for evidence that they have understood a subject.
What writing method does the post recommend?
Buffalo advises producing a rough first draft without trying to polish it as it is written. He also suggests keeping a working notes file to record ideas and questions during research.
Source: muratbuffalo.blogspot.com
