DeepSeek & ChatGPT BaZi Limits: Why They Fall Short

More and more people are sharing their experiences of using ChatGPT or DeepSeek for 'fortune telling' on social media. Type in your birth time, and within seconds you get a personality, career, and relationship analysis — it's a novel experience that has sparked genuine interest in BaZi (Chinese Four Pillar astrology).

As a team focused on BaZi analysis technology, we welcome this trend. Anything that introduces more people to traditional Chinese metaphysics is a positive development. But through practical use, we've also found that general-purpose large language models have certain structural limitations when it comes to BaZi analysis. This article offers an objective look at what generic AI can and cannot do for fortune telling, and what a truly professional BaZi system requires.

What ChatGPT and DeepSeek Do Well

To be fair, using general-purpose AI to explore Chinese metaphysics has real advantages.

First, it's free and convenient. Open a chat window, type in your birth time, and you're off. No apps to download, no specialized platforms to register for. For someone who just wants a casual introduction to BaZi, the barrier to entry is essentially zero.

Second, there's flexible conversation. You can ask follow-up questions like "What is an Indirect Seal?" or "What does Eating God controlling Seven Killings mean?" and the AI will explain in plain language. This interactive learning experience is something traditional BaZi books and static chart tools simply cannot offer.

Third, concept explanation is strong. General-purpose AI has been trained on extensive metaphysics-related texts and has a reasonable grasp of Five Element relationships, Ten Gods meanings, and basic pattern concepts, delivering highly readable explanations.

These advantages are real. If you're simply curious and want a rough overview of your chart, ChatGPT or DeepSeek will serve you well. But when you need an analysis you can seriously rely on, the picture changes.

Five Blind Spots of General-Purpose AI

Blind Spot 1: No True Solar Time Calculation

The first step in BaZi analysis is determining the birth hour, which requires true solar time — adjusting standard clock time based on the birth location's longitude. Being born at 2:00 PM in Beijing versus 2:00 PM in Chengdu can mean a difference of nearly an hour in true solar time, and that difference can change the Hour Pillar entirely, affecting the whole chart.

General-purpose AI doesn't make this calculation. You say 2:00 PM, it uses 2:00 PM, completely ignoring geographic time correction. For anyone born near an hour boundary, this means the entire analysis may start from an incorrect foundation.

Blind Spot 2: Pattern Classification by Guesswork

Pattern classification is the heart of BaZi analysis. Correct classification requires strict rules: examine the Month Branch first, check stem transparency, assess whether the Useful God is compromised, and finally determine the pattern type by weighing Day Master strength. *Zi Ping Zhen Quan* provides a detailed rule system for this process.

General-purpose AI has no built-in rule engine for this. Its pattern judgments are essentially pattern matching — learning from training data that 'this combination of stems and branches is often called X pattern' and making approximate inferences. This might coincidentally work for common patterns (like Direct Officer or Eating God patterns), but frequently fails for complex ones (Follow patterns, Transformation patterns, Mixed Qi patterns, etc.).

Blind Spot 3: Unreliable Classical Citations

General-purpose AI sometimes cites classical texts like *Zi Ping Zhen Quan*, *Di Tian Sui*, or *Qiong Tong Bao Jian*, which looks impressively scholarly. But careful verification reveals that a significant portion of these 'citations' are fabricated — the AI might mix content from different texts, or generate passages that simply don't exist in any source.

This is a well-known LLM limitation: they excel at generating plausible-sounding text but don't actually 'consult' any references. When users see citation-heavy analysis, it can create misplaced trust.

Blind Spot 4: Stem-Branch Hallucination

Large language models are notoriously unreliable at deriving year stems-branches and day stems-branches — this is a widely acknowledged issue in the field. The AI might tell you the Year Pillar for 1995 is "Jia Xu" (when it should be Yi Hai), or get your Day Pillar completely wrong. The Day Pillar is the most critical pillar in BaZi — the Day Stem is the Day Master, determining the foundation of all Ten Gods relationships. If the Day Pillar is wrong, every subsequent analysis becomes meaningless.

This isn't a flaw of any particular model but a structural problem across all general-purpose LLMs: calendar derivation requires deterministic algorithms, and probabilistic text generation models are fundamentally unsuited for this kind of precise computation.

Blind Spot 5: No Spirit Star or Branch Interaction Computation

Professional BaZi analysis requires computing spirit stars (Heavenly Noble, Literary Star, Blade of Sheep, Canopy, and 38 classical stars in total) and branch interactions (Three Combinations, Six Combinations, Clashes, Punishments, Harms, Destructions, Hidden Combinations, and more). Each has very specific derivation rules: for instance, the Heavenly Noble is derived by looking up specific branches based on the Day Stem, and a Three Combination requires three specific branches to be present with sufficient strength.

General-purpose AI cannot execute these rule-based derivations. It might know what "Heavenly Noble" means conceptually and explain its significance, but it cannot accurately determine whether your chart contains one, which pillar it appears in, or how it interacts with other stars.

What Professional BaZi Analysis Requires

A truly professional BaZi analysis system needs a deterministic rule engine to handle all computation, with AI responsible only for interpretation based on accurate data.

This technical stack includes at minimum:

Stem-Branch Calculation Engine — Based on astronomical calendar data, precisely deriving all Four Pillar stems and branches. Year Pillars use the Start of Spring as the boundary, Month Pillars follow the 24 solar terms, and Day and Hour Pillars are based on precise date-time calculation. All data is obtained through lookup tables with zero guesswork.

Ten Gods Derivation System — Automatically computing all Ten Gods based on the Day Master's Five Element relationships with every other character (including hidden stems). Direct Officer, Seven Killings, Direct Seal, Indirect Seal, Friend, Rob Wealth, Eating God, Hurting Officer, Direct Wealth, Indirect Wealth — each derived through deterministic rules.

Pattern Classification Module — Strictly following *Zi Ping Zhen Quan*'s methodology, step by step: assessing Month Branch stem transparency, evaluating Useful God status, determining pattern success or failure. This is a multi-step rule-based reasoning process that cannot rely on pattern matching.

Spirit Star Computation — Complete derivation rules for 38 classical spirit stars, each verified against original classical sources.

Qiong Tong Bao Jian Seasonal Table — 120+ entries mapping Day Stem to Month Branch for precise seasonal adjustment guidance.

Hidden Stems — The heavenly stems contained within each earthly branch and their relative strength proportions, directly affecting Ten Gods distribution and strength assessment.

Branch Interaction Analysis — Complete computation of Three Combinations, Six Combinations, Three Meetings, Clashes, Punishments, Harms, Destructions, Hidden Combinations, Hidden Three Meetings, Arch Combinations, and all other interaction relationships.

All of the above must be precisely computed by the rule engine, with structured data then passed to the AI model for interpretation. This is the core principle of the "rule engine + AI" architecture: let algorithms handle what's deterministic, let AI handle what requires understanding.

A Real Comparison

Consider someone born on November 18, 1995 at 8:58 AM, male. Let's examine how generic AI and a professional system would differ.

At the chart calculation level, generic AI might incorrectly derive the Day Pillar — Day Pillar calculation involves complex calendar conversion, and LLMs are extremely prone to hallucination at this step. A professional system derives all Four Pillar stems and branches through lookup tables and algorithms with zero possibility of guesswork.

At the pattern classification level, this chart actually features a Blade of Sheep pattern (Yang Ren Ge) — Rob Wealth dominates in the Month Branch, forming a special pattern structure. Generic AI would likely miss the Blade of Sheep classification criteria and produce either a vague 'somewhat weak/somewhat strong' assessment or an incorrect pattern type.

At the spirit star identification level, a professional system precisely computes specific stars like Heavenly Noble and Literary Star along with their pillar positions. Generic AI might offer vague statements like 'you may have noble person stars' without providing a definitive list.

At the branch interaction level, a professional system analyzes all relationships between earthly branches — combinations, clashes, punishments, harms — and their impact on the overall chart structure. Generic AI typically skips this analytical dimension entirely.

At the luck pillar timing level, a professional system precisely calculates the starting age and exact date range for each luck pillar period, accurate to the month. Generic AI might offer a rough luck pillar estimate, but both the timing and the pillar stems-branches may contain errors.

When Generic AI Is Good Enough

We won't claim 'you must use a professional system' — that wouldn't be honest.

If you're simply curious and want a general sense of what BaZi is about, what your Five Element belongs to, or what the Ten Gods represent, ChatGPT or DeepSeek is perfectly adequate. Generic AI does an excellent job with concept explanation and introductory education, often providing a friendlier experience than many traditional metaphysics websites.

If you want to learn BaZi basics, generic AI makes an excellent study companion. You can ask any beginner question and receive patient, plain-language explanations.

If you want an entertaining experience — chatting with AI about your personality and fortune as a form of self-exploration — generic AI handles this well too.

But if you care about analytical accuracy — perhaps you're consulting BaZi for major life decisions, or you have enough background knowledge to want a deep understanding of your pattern and fortune trajectory — then you need a system with 100% precise chart calculation, rule-based pattern classification, and complete spirit star and branch interaction analysis. This isn't a question of 'better' but of "whether it can be done at all."

FAQ

Is DeepSeek or ChatGPT fortune telling actually accurate?

Generic AI performs reasonably well at explaining BaZi concepts but has structural deficiencies in chart calculation. They may get the Day Pillar wrong, ignore true solar time correction, or misclassify the pattern type. For understanding basic concepts, the accuracy is acceptable; for precise chart analysis, a professional system is recommended.

Why do generic AI models get stems and branches wrong?

Large language models are probabilistic text generation systems that aren't designed for precise mathematical and calendar calculations. Year stem-branches require considering the Start of Spring boundary (not January 1st), and day stem-branches require complex calendar conversion — all of which need deterministic algorithms rather than probabilistic inference. This is a universal limitation across all LLMs regardless of model quality.

How is Deep Oracle different from using ChatGPT directly?

Deep Oracle uses a dual-layer "rule engine + AI" architecture: all chart calculations (Four Pillar stems and branches, Ten Gods, pattern classification, 38 spirit stars, branch interactions) are performed by a deterministic rule engine, with AI responsible only for deep interpretation based on accurate data. Using ChatGPT directly means both calculation and interpretation are handled by the same probabilistic model, with no guarantee of computational accuracy.

Are there privacy risks with using DeepSeek for fortune telling?

Using any AI tool involves data transmission. Privacy policies vary across generic AI platforms, so we recommend reading their terms carefully. Deep Oracle encrypts user birth information and does not use raw data for model training or share it with third parties.

Will generic AI eventually become professional enough?

Large language models continue to improve, but the chart calculation problem isn't something that 'stronger models' can solve — it's a fundamental gap between probabilistic models and deterministic computation. Even the most advanced models are unsuited for precise calendar derivation. The future direction is more likely what Deep Oracle already does: specialized rule engines for calculation, AI for interpretation.

I've seen people on social media get seemingly accurate results from AI fortune telling. How is that possible?

Some aspects of BaZi analysis (like Five Element attributes and general personality traits) may appear 'accurate' even with minor chart errors — this is similar to the Barnum effect. The real test of accuracy lies in precise dimensions like pattern classification, Useful God analysis, and luck pillar trajectories. We recommend verifying analysis quality using benchmark cases (such as people you know very well).

Further Reading

If you're interested in AI fortune telling and BaZi analysis, these articles are worth exploring:

AI Fortune Telling Accuracy Analysis — An in-depth look at the accuracy boundaries of AI metaphysics analysis.

How AI Analyzes BaZi — Technical details of the rule engine + AI architecture.

Complete Guide to AI Fortune Telling — From beginner to advanced: a comprehensive guide to AI-powered metaphysics.

Free Online Chart Calculator — Experience professional-grade BaZi charting and analysis.

Complete Guide to BaZi Patterns — A systematic overview of all BaZi pattern types.

Try Professional AI BaZi Analysis Now →

View Celebrity BaZi Case Studies →