The new ChatGPT button turns a chat into a looping GIF, and the number of images OpenAI ships each week has climbed from three billion to four billion in two months. The question the tool raises is why the thing the model makes to leave the app grows faster than the conversation that made it.
The model’s announcement describes it as sharper, with better text rendering. The GIF tool’s description describes what the model now has to do: pump out a shareable object at volume. Their disagreement is not about the engine but about value: the market quotes the announcement’s facts, resolution, and text rendering, while the volume figure of three billion to four billion in two months measures an object that arrives already formatted to be forwarded.
Tom Standage’s Writing on the Wall makes the mechanism concrete. On some platforms, more than 80 percent of items shared are repins or reblogs of items previously posted by other users, and the mere act of sharing can be a form of self-expression. The GIF is the extreme case of a format that is already a shared object before anyone taps. The model’s fidelity is the story the market ignores; the volume is the story the market moves on.
The common assumption is that the news is about a better model. What is actually happening is that ChatGPT has started manufacturing objects people want to send rather than things people want to read.
The GIF is the first output that needs no forwarding work: its loop carries its meaning, and its brevity means it arrives with nothing to add, so it travels without commentary. The model’s sharpest new feature is that it is now a source of handoffs, and the volume swing is the evidence, not a quality benchmark. The chat is the place where the model thinks; the handoff is the place where it stops.
That means the exit becomes the product and the model a cheap input; the shareable object is the sale. The weekly image volume, three billion to four billion in two months, is the unit of value, not the resolution. Every frontier lab faces the same choice; the winners will be the ones whose products are the best exits, not the best generators. A chatbot that shipped no shareable-output layer would probably look as quaint as a chat program that did not; the model would be interchangeable, and the toll would be collected at the layer where the object leaves the screen.
The GIF is the first output the model was built to hand off, and it already moves what the model holds. When the exit wins, the value lives in the leaving, not the making.
Leave a Reply