> In the present context, the fallacy manifests in claims that LLMs could not possibly be good models of some cognitive capacity because their operations merely consist in a collection of statistical calculations, or linear algebra operations, or next-token predictions
Nobody actually makes this argument though.