ChatGPT Images 2.0 adds unprompted "WHY ARE YOU LIKE THIS" sign to chaotic generation — Willison calls it a "glimpse of AGI"

Developer Simon Willison flagged an unexpected behavior in ChatGPT Images 2.0: when prompted to generate "a horse riding an astronaut, where the astronaut is riding a pelican that is riding a bicycle," the model spontaneously inserted a background sign reading "WHY ARE YOU LIKE THIS" — a phrase never mentioned in the prompt. Willison confirmed via independent testing that the sign was an entirely model-generated addition.

The incident has quickly become a talking point in AI circles as a rare example of a generative image model producing what reads as self-aware, sardonic meta-commentary on its own input. While far from a formal benchmark, it illustrates how frontier image models are developing unexpected emergent behaviors — injecting tonal or contextual editorial judgment that goes well beyond pixel-level instruction-following.