Mathematicians Just Proved How to Prevent AI Brain Rot

Main Argument: Preventing Cognitive Dependence on LLMs

Cal Newport discusses a paper titled “Big Language Models as a Cognitive Virus,” written by applied mathematicians associated with biophysics and complexity theory. The paper uses a mathematical population model to explore how LLM use might spread and affect people’s ability to think independently.

Newport emphasizes that the model is a way to reason about possible dynamics—not an empirical demonstration that its equations or parameter values precisely describe society.

How the Model Works

The model divides the population into three groups:

• U: People who do not use LLMs.
• C: People who use LLMs but retain independent thinking.
• D: People who have become dependent on LLMs and have largely lost the ability to think independently.

A system of differential equations estimates how the proportions in these groups change over time. Researchers set parameters representing factors such as:

• Pressure to begin using an LLM.
• The rate at which users become dependent.
• The chance that dependent users regain independent habits or return to non-use.
• Social pressure from non-users.

The model is repeatedly updated to see whether the population settles into a stable balance among the groups.

The Model’s Main Findings

• A tipping point may occur: If pressure to use LLMs is low, most people remain non-users. Once pressure passes a threshold, LLM use can become widespread.
• The effects may be hard to reverse: The model displays hysteresis: after LLM use becomes common, pressure may need to fall substantially before usage returns to its earlier level.
• Prevention may be easier than reversal: If dependence is the outcome to avoid, it may be more effective to prevent people from becoming dependent than to try to restore independent habits later.

The paper’s concern is not simply that people use LLMs. It is that many people could shift into D—using them in ways that erode their ability to think for themselves.

Three Possible Interventions

1. Reduce pressure to use LLMs. This can keep more people in U. The paper’s authors consider this difficult to achieve broadly because LLMs are useful and increasingly widespread. Newport argues it can still make sense in particular settings:

   • Universities may restrict LLM use where the goal is to develop students’ minds.
   • Publications may require reporting and writing to be done by humans, protecting the value of human judgment and prose.

2. Encourage people to return to non-use. The model includes a force that moves users back toward U. Newport points to changing cultural attitudes as one way this might happen. He compares it to campaigns that helped make smoking less appealing to young people, including the Truth campaign, which framed tobacco companies as exploiting them.

3. Keep users from moving from C to D. This means preserving independent thought while allowing people to use LLMs. The paper’s authors favor this intervention: rather than trying to stop LLM use, societies and organizations should protect activities that require human contemplation and thinking.

Newport also favors the third approach because individuals can begin applying it immediately. He describes this as part of a broader “cognitive fitness” idea: treating the ability to think deeply as something to maintain and train, much like physical health. He still sees the first two interventions as appropriate in certain organizational or cultural contexts.

Newport’s Three Practical Rules for Using LLMs

• Automate boredom, not cognitive friction. Using an LLM for a dull task may be reasonable. Avoid using it to skip the difficult thinking, problem-solving, or creative work that helps build skill.
• Don’t pass off LLM-generated writing as your own. Newport recommends clearly labeling text produced by an LLM. Writing yourself forces you to clarify and express your thoughts; outsourcing it can weaken that practice and mislead readers about who wrote the text.
• Prefer human writing to extra layers of AI summary. Books and articles are already the author’s chosen compression of a subject. Newport cautions against summarizing them again merely to avoid the effort of reading. He sees summaries as useful when a document is very long and only a small portion is relevant, or when a communication is needlessly verbose.

Newport’s Response to Criticism of His AI Views

Newport addresses a Jesse Singal article in The Dispatch that characterized him as a leading “anti-doomer.” Newport says the debate about AI is shaped in part by futurist beliefs associated with rationalism and parts of effective altruism. In his account, some proponents expect superintelligent AI to bring either an extraordinary utopia or catastrophe, depending on whether it is developed and controlled correctly.

His concern is that this outlook may affect how people discuss AI and what companies choose to build or test. He argues that observers should compare the claims of labs influenced by those beliefs with the views of researchers and companies working on similar technology without the same ideological background. He points to Mark Zuckerberg and Jensen Huang as examples of industry figures who have criticized some claims or priorities, while acknowledging that they have their own interests and biases.

Newport also argues that beliefs about recursive self-improvement can influence research priorities and risk-taking. He says understanding these ideologies is important for interpreting the AI debate—not because the people involved are unusual, but because their beliefs may shape their decisions. He identifies himself as critical of “doomer” narratives, while also expressing concern about how ideology may influence leading AI labs.

Other Topics

• Vision documents: Newport reviews examples of one-page plans that identify a person’s main goals, projects, and the tools and routines they will use to make progress. He recommends keeping such a document somewhere visible and using it to stay oriented amid daily distractions.
• Time blocking and parenting: Asked whether a parent caring for an eight-month-old should time-block the day, Newport says time blocking was designed mainly to help knowledge workers manage overload and protect important work. He advises supporting the parent rather than imposing a schedule on unpredictable childcare.
• Book discussion: Newport mentions a book about simplification by Axios founders Jim VandeHei, Mike Allen, and Roy Schwartz. He appreciates its emphasis on simpler priorities and reducing low-value work, though he notes that the authors recommend LLM use for writing more than he does.
• Advertising: The episode includes sponsor messages for Factor, Caldera Lab, Notion, and MyBodyTutor.

Speakers and Sources Featured

• Cal Newport — host and principal speaker.
• Jesse — off-camera production collaborator addressed during the episode; the subtitles do not provide a surname.
• “Big Language Models as a Cognitive Virus” — paper discussed by Newport; its authors are described as applied mathematicians, but they are not named in the subtitles.
• Jesse Singal — author of a Dispatch article Newport discusses.
• Other people cited in the AI discussion: Mark Zuckerberg, Jensen Huang, Timnit Gebru, Eliezer Yudkowsky, Nick Bostrom, Dario Amodei, Greg Brockman, and Ed Zitron.
• Other sources and examples mentioned: Epidemiological population models, the Truth anti-smoking campaign, Newport’s New York Times writing on cognitive fitness, and a book on simplification by Jim VandeHei, Mike Allen, and Roy Schwartz.
• Sponsors featured: Factor, Caldera Lab, Notion, and MyBodyTutor.