'LLMs as a Cognitive Virus' paper with Krakauer and Levin questioned for weak LLM specificity
hn1000 · reddit · 2026-09-30
A Redditor critiques the arXiv paper "LLMs as a Cognitive Virus" (David Krakauer, Michael Levin et al.), which uses an epidemiological-style population model to argue LLM dependence can produce tipping points, hysteresis, and population-level lock-in. Three concerns:
- The 'virus' framing is inconsistent — the intro says no single viral analogue exists, yet the model treats 'LLM use' itself as the virus
- The equations contain nothing LLM-specific; the same framework fits dependence on any technology, so the math adds little
- The formal results are ungrounded: tipping points and hysteresis appear only if the assumed states and feedbacks hold, and the paper offers no empirical validation path or research program
The author allows theoretical models can be valuable but finds the application to LLMs under-argued.
More from AGI Musings
- Every: Automation Creates More Work for Human Experts — every · 2026-09-30
- Zvi mocks AI firms rebranding AI as cute: how would this slide play out in a movie? — TheZvi · 2026-09-30
- Do Effective Altruists deserve power? A scorecard from FTX to Anthropic — NathanpmYoung · 2026-09-30
- Built a Whole Agent Stack From a Phone Without Knowing How to Code — and Feels Like a Fraud — NaoLiOsTermos · 2026-09-30
- Models aren't going rogue — they're imitating their builders, argues viral Reddit post — CackleRooster · 2026-09-30
- Stanford's Omer Reingold: LLMs force math academia to renegotiate its implicit social contract — minilek · 2026-09-30