Large language models now assist across language, code, and multimodal workflows, but the banner of emergence can blur limits and inflate expectations. We argue for precise capability claims, transparent evaluation, staged releases, and governance matched to demonstrated risks and trade offs.