Why LLMs Get Astrology Charts Wrong — and Never Admit It

Large language models are increasingly being used to generate Korean saju (Four Pillars) and Chinese BaZi birth charts, but a developer running a saju reading service warns the results are structurally unreliable. Generating an accurate chart requires precise astronomical calculations — including solar term instants, sexagenary cycle counting, and true solar time corrections — none of which LLMs actually perform. Instead, models predict what a chart typically looks like for a given date, which works for common cases but silently fails at boundary conditions such as births near solar term transitions or late-night hour conventions. The deeper problem, the author argues, is that LLMs produce wrong answers with the same confident tone as correct ones, offering no indication of uncertainty. Their proposed fix is a strict separation of roles: a deterministic open-source engine handles all calendar arithmetic, while the LLM is only permitted to write the interpretive prose based on pre-computed inputs.