Is Fortune-Telling With DeepSeek or ChatGPT Accurate? Why AI Gives a Different Answer Every Time

Over the past year, "AI fortune-telling" has gone fully viral on social media. All kinds of "metaphysics prompt" templates circulate: turn on DeepSeek's deep-thinking mode, paste in the prompt, swap in your own birth date and birthplace, and a few minutes later you receive a sprawling, detailed reading.

Some people are amazed at how closely the result matches what an expensive in-person master once told them. But more and more are running into an awkward problem: the same person, the same prompt, tested twice, produces two readings that don't agree.

As a metaphysics platform built by AI engineers, we want to answer this honestly: it isn't your imagination — it's a technical inevitability.

Problem One: Language Models Are Built to Vary

Products like DeepSeek and ChatGPT are powered by large language models (LLMs). They work by predicting, word by word, "the most likely next token" given what came before — and that prediction process deliberately keeps some randomness (technically called *temperature*), so the conversation feels natural rather than robotic.

That design is a strength for chatting and a fatal flaw for fortune-telling. Any reading has to start with calculating the chart: converting the birth time into the sexagenary (stem-branch) system, fixing the Life Palace, placing the fourteen major stars, laying out the Four Transformations — every step is a deterministic rule computation, and the same input *must* yield the same chart. A language model has no built-in calendar-and-rules engine; it is "reciting from memory" the metaphysics texts it saw in training. Memory drifts, so the results drift.

Problem Two: When the AI Doesn't Know, It Invents

Language models have a well-known weakness: hallucination. When their training data lacks reliable knowledge for a question, they don't say "I don't know" — they generate something that *sounds* perfectly reasoned.

Metaphysics is a disaster zone for hallucination. The metaphysics texts online are wildly uneven in quality and mixed across schools, so the model learns one big muddled stew. The result: you'll see the AI place a Zi Wei star in the wrong palace, miscalculate a BaZi Ten God, and then wrap the error in fluent, confident prose that makes it seamless. Someone unfamiliar with the craft has no way to catch it.

So Is AI Fortune-Telling a False Proposition?

No. The problem isn't AI — it's the architecture.

The right approach separates calculating the chart from interpreting it. Hand the calculation to a deterministic rules engine — strict calendar conversion and star-placement rules faithful to the classical texts, so the same birth time yields the same chart ten thousand times over. The AI only handles the second step: translating an already-calculated, correct chart into language a modern reader can understand.

That is exactly Xuan Zhang's architecture. Zi Wei charts are computed with the open-source engine iztro, running deterministically according to Dou Shu rules handed down since the Tang and Song dynasties; palm analysis uses computer vision to measure the lines objectively. In our system the AI is a *translator*, not a *diviner* — it interprets the chart but never takes part in calculating it. So whether you test today or next month, the chart is the same chart and the conclusion is the same conclusion.

When Is It Fine to Use a Chatbot for Readings?

To be fair: if you're just having fun and want to see how the AI describes your "past lives," DeepSeek and ChatGPT are more than enough — and free. In an entertainment mindset, accuracy was never the point.

But if you seriously want to use metaphysics as a framework for understanding yourself — a reading you can re-verify, one that holds up against a second look — then "a different answer every time" isn't a minor blemish; it's a fundamental defect. An answer that drifts can't serve as the basis for any decision.

Closing

The rise of AI fortune-telling proves one thing: what people want was never just "accurate," but a self-knowledge tool that is stable, consistent, and re-checkable at any time. A general chatbot can't give you that. The right architecture can.

Want to feel the difference between "engine calculates, AI interprets"? Free AI palm scan — results in 60 seconds, and you're welcome to test it twice and compare.

---

*The metaphysical content in this article is for reference only. For a detailed comparison of AI palm reading versus traditional palmistry, see this article.*