AI and the sky

AI Hallucinations: When the Machine Makes Things Up

An AI hallucination is a fluent, confident answer from a language model that is false or invented. The model generates plausible text rather than checking facts, so for planet positions, eclipse dates and sources it is safest to rely on calculation and primary publications.

Anyone who has chatted with an AI assistant has met the moment when an answer sounds detailed, calm and convincing, and then turns out to be wrong. This is called an AI hallucination. Here we explain what it means, how it can appear in questions about the sky, and and how Astronativa guards against it.

What a hallucination is

A large language model (LLM) is a program trained on a vast amount of text. Its core job is to continue a piece of text in the most plausible way. It does not hold facts the way an encyclopedia or database does; it has learned how words tend to appear next to each other.

As a result it can produce a sentence that is smooth and grammatical yet does not match reality. That is a hallucination: a confident but incorrect or entirely made-up answer. The word is a metaphor. The model is not lying on purpose, because it has no intention to deceive. It simply does not separate “sounds right” from “is right.” AI developers and research institutions discuss this problem openly and work to reduce it, but nobody claims it has disappeared. For the bigger picture, see what AI can and cannot do.

Typical examples from the sky

Astronomy is full of numbers, dates and names, which is exactly where errors are sneaky:

• Wrong dates. A model may state an incorrect date for an eclipse, an equinox or a meteor shower peak, because dates like these appear often in its training text and it guesses a plausible one.
• A wrong planet position. Asked which sign Mars is in on a given day, a model may answer with confidence without calculating anything. A planet’s position is the result of a computation, not a remembered sentence.
• Invented sources. Sometimes a model cites an article, an author or a web address that sounds realistic but does not exist. This is one of the best-known forms of hallucination.
• Blended concepts. An astronomical fact and an astrological claim can merge into one sentence, such as “Mercury retrograde causes problems.” Retrograde motion is an apparent motion caused by perspective, as Earth and the other planet move along their orbits at different speeds; linking it to events belongs to astrological tradition and is not an established fact. Read more in astronomy vs. astrology.
• False precision. A model can give a number with many decimals that looks scientific but is made up.

Why Astronativa does not “ask” AI for positions

The positions of the Sun, the Moon and the planets are not a matter of opinion or memory. They are calculated. Astronomers use ephemerides (tables or mathematical models giving the positions of celestial bodies at any moment). Scientific institutions such as NASA’s JPL publish this kind of data, and its Solar System Dynamics resources live at ssd.jpl.nasa.gov.

So Astronativa follows a simple principle: numbers come from calculation, words come from explanation. Planet positions for the 3D sky and for birth charts are computed with ephemerides and astronomical formulas. When AI is involved, it helps explain, translate or retell what has already been calculated; it does not invent the positions. Details are in ephemerides and accuracy and calculation versus narrative.

Think of the difference between a calculator and a storyteller. The calculator gives the same correct result every time; the storyteller may embellish. Both are useful, for different jobs.

How to spot and check a doubtful answer

A few simple habits go a long way:

1. Do not trust the tone. Confidence is not evidence. Smooth text can be wrong.
2. Find a second source. Check an eclipse date or a Moon phase on a NASA or ESA page, or on the site of another scientific institution.
3. Verify citations. If an answer names a book, an author or a link, open it. If it does not exist or does not say what is claimed, do not use it.
4. Be wary of very precise numbers. Ask where they come from. If there is no clear origin, treat them with caution.
5. Separate fact from reading. Ask: is this a measurement, or an interpretation from a tradition? How to tell fact from interpretation can help.
6. Ask better questions. A specific question with date, time and place leaves less room for guessing. See how to ask AI about the sky.
7. Allow “I don’t know.” Ask the model explicitly to say so when it is unsure.

For a wider framework, see AI fact-checking.

A calm attitude

Hallucinations do not make AI useless. It is good at explaining, summarizing and translating, as long as we do not make it the source of measured data. The healthiest approach is to treat it like a diligent assistant: listen with interest, but verify the important numbers. The sky stays where it is, and we can always compare it with science, observation and a little patience.

Positions in Astronativa are astronomically exact. The interpretation is for inspiration. · Glossary · Terms · Privacy