Is this OpenAI catching up with Anthropic artifacts? At the same time they say "We’ve trained GPT‑6 to compose responses using text, visuals and interactive elements...", rather than a harness.