I read the title and thought to myself, "The chance that this is a well-grounded, unbiased, and robust design is close to 0." So I started reading...
And having read it, what a terrible piece of science slop this is!
There is no observed adoption dataset used to establish that this mechanism explains LLM use better than alternatives such as usefulness, falling prices, workplace requirements, or independent discovery. The behavior they talk about could describe many socially adopted technologies or practices. Its applicability does not identify anything distinctively viral or harmful about LLMs.
The cognitive decline is assigned, not measured or independently derived. They calculate population competence by averaging those assigned scores across the groups. Consequently, moving people into lower-scoring groups lowers the average by construction. The dramatic fall from 1 to approximately 0.425 in their example is a model output conditional on those choices, not evidence of a 57% real-world cognitive decline.
They explicitly define competence as what the human can do when external support is withdrawn, excluding the total capability of the human/AI system. This cannot settle whether the technology improves overall performance or human welfare. And they provide no measurement protocol clarifying exactly which supports could be removed for a fair comparison (for instance, how does this compare to Wikipedia or electricity). They don’t do this because of an agenda, so allow me to assert something with just as little evidence as the authors provide for their conclusion: This is obvious bias, and they are designing the study to get the result they want, and they don’t care about scientific truth, just about getting their paper published and attracting attention.
And the “runaway” stuff is just not supported by the model. The language about equilibrium as a parameter changes is not automatically evidence of an imminent, rapid societal collapse! it is PURE conspiracy-level rhetoric. Just put two things in the same sentence and hope people think they are related. This is misleading language at best, and willful manipulation at worst.
This is a BASIC failure of the minimal requirement for something to be called scientific. If there had been even a trace of a genuine attempt at Popperian falsification, the current research design would never have gotten past review. This is cargo cult science, it LOOKS like scientific research, but it’s not even close to scientific.
EVERYTHING I hate about modern research on full display! And I am losing respect for the HN community more and more, because only terrible studies with sensationalist claims and titles get attention. This is tabloid-level quality, just with a direct link to the article instead of a dumbed-down intermediate write-up. If anybody read the actual paper, this would not get posted ANYWHERE. But “LLMs are a cognitive virus” is what gets posted.....