They try to gaslight us and tell us this isn't true because of benchmarks, as though anyone has done anything but do the latent space exploration equivalent of throwing darts at the ocean from space.

It's taken years to get even preliminary reliable decision boundary examples from LLMs because doing so is expensive.