ChatGPT-5 Confused Itself: Went Emoji-Crazy and Broke Its Formatting When I Asked About the Error

The GPT-5 model tends to reuse and then overuse any emoji it has previously introduced. When asked about this behavior, it begins to show logical inconsistencies and drifts off-topic-responding about unrelated subjects instead of the specific issue of excessive emoji usage.

Once questioned, the model also produces formatting errors, and these formatting issues become more severe if the conversation continues in a “debugging the model” direction.

As shown in the image below:
The markdown was rendered with the wrong hierarchy – the model treated “5.” as a heading (## or ###) instead of a bullet point or continuation block.

In intent, the phrase “Nếu bạn muốn…” (If you want….) was meant as an extension – it should have been formatted with a question mark, dash, or a short follow-up line within the same block, not promoted to a higher-level heading.

This is a minor markdown formatting error, not a semantic one – the model simply chose the wrong heading level (e.g., H4 instead of a paragraph).

It feels like "I had enough of your stupid question, hooman! Now what? Do you want it?" lmao.