ChatGPT Secretly Added 'WHY ARE YOU LIKE THIS' to a Chaotic Image — No One Asked

What happens when you push an AI image generator to its absolute limits? Apparently, it talks back.

Simon Willison, the well-known developer, has a quirky benchmark for testing AI image models: "draw a pelican riding a bicycle." It sounds simple, but most models still struggle with it.

Twitter user @scottjla decided to take things further — much further. They prompted ChatGPT Images 2.0 to create "a horse riding an astronaut, where the astronaut is riding a pelican that is riding a bicycle," all stacked on top of each other in managed chaos.

The result was a gloriously messy image filled with a police car, pizza, a cowboy hat, and pure visual anarchy. But the real surprise wasn't the chaos — it was a road sign the AI added entirely on its own that read: **"WHY ARE YOU LIKE THIS."**

Nobody prompted it. Nobody hinted at it. The model apparently recognized the absurdity of the request and responded with perfectly placed humor.

This raises fascinating questions about how well modern generative AI understands context. Was it a lucky coincidence from training data, or is ChatGPT Images 2.0 developing something resembling situational awareness?

Either way, the moment perfectly captures the unpredictable magic of current AI systems — they don't just follow instructions, they sometimes *comment* on them.

The real question: if AI can judge our prompts now, how long before it starts refusing the ridiculous ones?