The New ChatGPT Is More Show Than Tell

OpenAI is rolling out what it calls Intelligent UI for ChatGPT, according to WIRED. Instead of answering only in text, ChatGPT can now generate custom visual elements when they help: interactive diagrams, calculators with sliders, clickable maps. The update runs on OpenAI's GPT-6 model and was set to reach paid users on Wednesday and free users the following day.

In WIRED's hands-on, ChatGPT produced a tappable anatomy diagram, an apartment affordability calculator, and a seat explorer comparing cabin layouts across several planes. OpenAI says its design team worked on when visuals add value versus clutter, and users can ask for fewer of them. WIRED notes the move resembles the generative UI Google has added to Search.

We think this is a real improvement for a lot of everyday questions. Some answers are just better as a picture or a tool. A budget question with sliders beats three paragraphs of hypotheticals. A diagram someone can poke at teaches faster than a description. Bringing that to a product WIRED describes as serving over a billion users could make AI help more understandable for people who find long text answers hard to parse.

The trade-off is subtler. When a chatbot writes text, readers can see its reasoning laid out, flaws and all. When it builds a polished interactive calculator, the assumptions are baked into the widget. Which costs were included? Which numbers were guessed? A slick interface can make a shaky answer look authoritative. We would like to see these generated tools show their inputs and sources clearly, especially for money, health and travel decisions.

There is also the design question WIRED raises. If the model increasingly decides what software looks like for each person, the craft of interface design shifts from building screens to setting rules for a system that builds them. That could free designers from repetitive work. It could also make products feel less coherent, and harder to audit, when every user sees something different.

Our read: a welcome change that makes ChatGPT more useful, with transparency as the thing to get right. I'm watching whether generated visuals clearly label their assumptions, and how often they get the underlying numbers wrong.

Chatbots have been moving toward multimodal experiences that blend text, images and voice. Google introduced a similar generative UI approach in Search earlier this year, so this is quickly becoming a competitive standard for AI assistants rather than a one-off feature.