This year’s college entrance exam: all five AI prediction models failed—but the essays they wrote were actually quite good.
“To know what you know and what you do not know, that is true knowledge.” — The Analects
The 2026 Gaokao essay topic has been released. Before the exam, five AI models (ChatGPT, Claude, Gemini, Doubao, and Kimi) all missed the prediction entirely. The Ministry of Education issued a warning against such hype. Yet, when these same AIs actually wrote the essays, their quality was surprisingly solid. Behind this contrast lies an important lesson for side hustles: you must clearly understand what AI can and cannot do.
Every year, as soon as the Gaokao essay topic is announced, everyone online rushes to write one.
This year was different—alongside the 12.9 million students, a swarm of AIs joined in.
And here’s the twist: the AIs completely failed at predicting the topic—but their actual essays were quite impressive.
If you’re using AI for your side hustle, this contrast deserves careful thought.
📉 AI Prediction: All Five Models Failed
Before the exam, several media outlets used five major AI models—ChatGPT, Claude, Gemini, Doubao, and Kimi—to predict the Gaokao essay topic.
What happened? All of them missed.
Each model independently predicted a theme centered on “technology and ethics”: ChatGPT predicted “scale,” Claude predicted “the scenery on the road,” Gemini predicted “frictionless living,” Doubao predicted “the era of shortcuts,” and Kimi predicted “algorithmic navigation.”
The actual exam question? “How has the meaning of words changed over time? Which word’s meaning has shifted for you? Discuss your personal growth.”
It had nothing to do with technology ethics—completely opposite in direction.
On the same day, the Ministry of Education issued a public warning: using AI to predict exam topics constitutes false advertising and fraud. The exam design intentionally avoids predictable patterns and clichés.
📌 AI can write—but it cannot predict. This is the first boundary.
✍️ AI Takes the Exam: Quality Is Solid
While the predictions failed, what happened when these AIs actually wrote the essay?
Three AIs—DeepSeek, Qwen, and Doubao—wrote responses to the same prompt. The results surprised many.
For Beijing’s micro-essay prompt—“Write a promotional message for an elderly care home about AI and a happy retirement”—all three produced copy that was nuanced, appropriately toned, and flawless.
For Shanghai’s main essay prompt—“As technology transforms the world, it also reshapes our imagination”—each AI offered a distinct perspective:
DeepSeek: “Technology is the catalyst for imagination’s rebirth through fire.”
Qwen: “Beware of human thought being trained by algorithms.”
Doubao: “Use technology to expand the boundaries of cognition.”
Each response was insightful, well-reasoned, and stylistically strong. If submitted in an actual exam, they would have earned at least an above-average score.
AI can’t predict—but it can execute instructions well. This is the second boundary.
💡 Implications for Side Hustles
These two boundaries map directly to two critical questions for side hustlers:
AI cannot make decisions for you (like failing to predict the topic)
It cannot choose your niche—whether to write copy or build agents, you must decide
It cannot judge timing—whether to chase a trend is your call
It cannot take responsibility—if a client is unhappy, you can’t blame “the AI”
AI can help you produce work (like writing strong essays)
You give it direction—it drafts
You give it structure—it fills in content
You give it tone—it follows
The biggest mistake? Letting AI choose your direction (predicting) while you manually produce the output (writing). The correct approach is the reverse: you choose the direction; AI executes.
📌 The most accurate description of AI is not “an all-purpose substitute,” but a tireless executor—whose direction must always be set by you.
#AI #AIAgent @grok
