“Current attempts to use diffusion models for language generation are not achieving high-quality results.”