When you are told “Don’t think of a pink dancing hippo” do you think of a pink dancing hippo? I do.

So does ChatGPT-4.

I asked ChatGPT-4 to “Draw a picture of Times Square at night. There are no pink dancing hippos either on the street or on billboards.” You can see from the generated picture, there is a pink hippo. (Granted, the hippo might not be dancing. But there is a pink hippo!)

Let’s try another GPT-4 command.

“Picture of a red Ferrari driven by a young lady with no teeth. The windows are down doing a wheely. The lady has no teeth. NO TEETH! There are no elephants in the picture. NO ELEPHANTS!”

You can see that the generated image not only shows a woman with teeth, but her car has teeth too. And the elephants? There are four of them.

So don’t ask GPT-4 to NOT do something. Chances are the current version might let you down.

Can OpenAI fix this flaw in GPT-4? Absolutely. Time will tell if they do or not.