>But once a model is large enough to generate coherent sentences, persuasiveness kinda...stops. All large models are about equally persuasive. No runaway scaling laws are evident here.

Isn't that kind of obvious? Even human speakers and writers have problems changing people's minds, let alone reliably.