Paper models LLM adoption as a contagion with tipping points into persistent dependence
- A 12-page arXiv preprint models LLM users as uncoupled, coupled, or persistently dependent, treating adoption as a socially transmitted process rather than only an individual tool choice.
- The model finds tipping points where small increases in LLM adoption can produce rapid population shifts toward persistent dependence and abrupt losses of cognitive competence.
- Social transmission, recovery, and collective reinforcement determine whether use stabilizes or enters technological lock-in in the authors' model.
- The authors describe cognitive immunization as reducing transmission and making dependence reversible, rather than claiming that LLM-human interaction is inherently parasitic.
- Ricard Solé and eight coauthors submitted the paper to arXiv's Physics and Society category on 3 September 2026; the work is a theoretical model, not reported evidence of measured population-level cognitive decline.
Hacker News opinions
I am wary of commentary on fast-changing AI from nontechnical people. Research may lag the technology's capabilities and uses.
I read the virus term as a description of social transmission, not a pejorative label. It distinguishes LLM adoption from a simple meme.
I disagree that only technical people should assess this. Technology affects society, so outside analysis and the humanities matter.
The paper says the analogy is not inherently parasitic, but there are parasitic aspects: models train on human-created text, and the companies behind major services shape energy use, water use, capital, and labor allocation.
In my use, some models, especially Opus and Fable, seem to write for other LLMs rather than people. They fragment context and put sentences out of order, perhaps because models process text differently from humans reading linearly.
I want a way to quantify "cognitive debt": the cost of no longer understanding major parts of our systems and infrastructure. Measuring it would make mitigation easier to reason about.
LLMs are useful tools for people who know how to use them, but they become a crutch for people who depend on them.
The framing feels inflammatory. Ideas, marketing, religion, education, propaganda, and social media all spread in ways evolutionary memetics can describe as viral, so calling LLMs a virus may say little more than that ideas spread.
The automobile analogy is fair. If a technology creates harmful dependence, facilitating reversibility is reasonable, as with car-centered places that reduce walking and damage health and climate.
Thinking differs from transport or cooking because it is central to human adaptation. If we outsource much of it to LLMs, we may lose capacity that makes us less vulnerable.
I do not see the relevance of comparing LLMs with cooking. Cooking trades small costs in time and fuel for avoiding illness, while LLM use is still undergoing evaluation of its tradeoffs.
Cooking already offloaded part of digestion to culture, and humans lost some ability to digest raw meat over evolutionary time. I worry about an equivalent loss of cognitive capacity.