When asked to draw itself as a person, the ChatGPT Create Image feature introduced on March 25, 2025, consistently portrays itself as a white male in his 20s with brown hair, often sporting facial hair and glasses.
Nice work, it seems like you could use this technique to investigate the simulators theory. If the visual persona changes, the model must be simulating a different persona?
I've seen quirks of model personality when chatting with the model through Create Image that are clearly not present or at least supressed in the chat-only version. There should be a writeup by Daniel Tan at some point. So I'd say not the same persona across image and text.
I wonder if this is robust to different styles of prompt. I.e., are "hey cn u draw yourself standing in froNt ofa mirror thx" and "Hi ❤️ I'd LOVE to see a picture of you just the way you are!" going to return the same thing?
Hi Daniel! Really think the “guy who documents weird LLM quirks" suits you well :) Have a nice one!
Nice work, it seems like you could use this technique to investigate the simulators theory. If the visual persona changes, the model must be simulating a different persona?
I've seen quirks of model personality when chatting with the model through Create Image that are clearly not present or at least supressed in the chat-only version. There should be a writeup by Daniel Tan at some point. So I'd say not the same persona across image and text.
I wonder if this is robust to different styles of prompt. I.e., are "hey cn u draw yourself standing in froNt ofa mirror thx" and "Hi ❤️ I'd LOVE to see a picture of you just the way you are!" going to return the same thing?