Why AI Saju (ChatGPT, Gemini) Keeps Getting It Wrong — The Ten-Thousand-Year Calendar Trap

These days you can ask a free AI for a Saju reading. But when the answer comes back, it often gets the most basic thing wrong — the eight characters of your Four Pillars. The longer someone has studied Saju, the more likely you are to hear them say, "AI Saju can't even get the chart right." Why is that? The answer is simple — a language model doesn't calculate your chart, it generates something that merely looks plausible.

Why does AI get even the chart wrong?

Because a language model generates a plausible-looking chart instead of calculating one. Tools like ChatGPT, Gemini, and Claude are models that stitch language patterns together by probability. They're superb at choosing the statistically likely next word, but the Four Pillars of Destiny aren't something you can arrive at that way. They're values you get by plugging a birth date and time into an ephemeris and calculating the solar terms and the stem-branch pairs (ganzhi). So when you feed a birthday into a language model, you frequently get a chart that reads convincingly but has the wrong characters.

How does AI Saju actually go wrong?

Long-time users tend to point to the same handful of failures.

• Sexagenary cycle errors: For example, some tools call 2026 a Bing-Yin (丙寅) year instead of the Bing-Wu (丙午) year it actually is. The Sixty Jiazi is a regular, repeating cycle, but when you're picking characters by probability, it's easy to lose the count.
• Wrong from the eight characters up: Especially on days where the solar-term boundary or the ganzhi count is tricky, it can't pin down the Year, Month, Day, and Hour Pillars correctly, and the Ten Gods (Companions, Output, Wealth, and so on) come out wrong too — so you often have to tell it the right answer before it finally falls in line. (It may nail ordinary dates, but near a boundary it's hard to trust.)
• A different answer every time: Ask about the same birthday again and both the chart and the reading change. Because this is probabilistic generation, every ask is a fresh roll of the dice.
• Even dedicated tools have limits: Some custom chatbots that brand themselves as Saju tools still need a separate external calculation bolted on just to handle the perpetual calendar properly.

Why can't AI get the Ten-Thousand-Year calendar right?

The heart of it is the difference between generating and calculating. The Ten-Thousand-Year (perpetual) calendar belongs to the realm of astronomy. The boundaries between solar terms are fixed by the position of the sun, and even on the same day the Hour Pillar can split depending on how you handle true solar time, daylight saving time, and the late-night Zi hour. This is minute-by-minute astronomical calculation, not something a "plausible sentence" can land. A language model doesn't have this calculation built in, so the closer a birth falls to a boundary — or the older a daylight-saving window it lands in — the more often it drifts off.

How is Gwiraedang different?

Gwiraedang doesn't generate your chart — it calculates it. Using astronomical calculation aligned with the standards of the Korea Astronomy and Space Science Institute (KASI) — solar terms down to the minute, with true-solar-time, daylight-saving, and Zi-hour corrections included — it sets the eight characters precisely. That's why the same birth date and time always yields the same chart, no matter when you ask (it's deterministic). Wherever a correction was applied, we show it transparently on screen. Because the interpretation built on top of a chart only means something if the chart itself is right.

An accurate chart is only ever a starting point. Saju isn't a tool that hands you a fixed fate — it's a reference for understanding yourself, meant for self-reflection, not fortune-telling. Two people with the same chart still live very different lives. If you'd like to read a chart yourself, see How to Read Saju; and if you're curious about the "Saju isn't prophecy" stance, Saju Is Not a Fixed Future makes a good companion read.

FAQ

Is it okay to get a Saju reading from ChatGPT? Asking for fun is fine, but if the chart itself is wrong, everything built on top of it wobbles right along with it. If you were born near a solar-term boundary or around the Zi hour, it can be off from the eight characters up. When you need an accurate chart, you're safer using a tool that does the astronomical calculation.

Why do I get a different answer every time I ask about the same chart? Because a language model generates by probability rather than calculating. Even for the identical question it rolls the dice again each time, so the chart and the reading shift. When a chart is built by astronomical calculation, the same birthday returns the same result whenever you ask.

What's the difference between AI Saju and a Ten-Thousand-Year calendar? The Ten-Thousand-Year calendar is a table that calculates the stem-branch pairs by plugging your birthday into an ephemeris. A proper Saju tool has this calculation built in and produces an accurate chart corrected for solar terms, true solar time, and daylight saving. An ordinary language model has no such calculation, so it's prone to answers that look plausible but are simply wrong.

So when you do turn to ChatGPT for a reading, the order of steps matters — How to Read Saju with ChatGPT walks you through a safe three-step approach.