Fel Conjecture

There is no reliable source for the term “Fel Conjecture.” Neither in mathematics nor in AI research can a recognized statement of this name be verified. It is possible that this is a typo, a very specific footnote, or a freely invented name. That is why this entry honestly explains what can be explained: what a conjecture in science actually is. And why an AI system is especially prone to getting such names wrong. Anyone who has encountered the term in a specific text should look there for a source citation.

Claim without proof: what defines a conjecture

In mathematics, a conjecture is a precisely formulated statement that no one has yet proven. It is more than a gut feeling. Usually, years of observation lie behind it: someone checks thousands of examples, and the statement holds true every time. Nevertheless, that does not count as proof. A proof must show that the statement holds for all cases, including those that no one has checked.

Famous examples are the Goldbach Conjecture and the Collatz Conjecture. Both are easy to understand and have remained unsolved for decades. Computers have confirmed them for enormous ranges of numbers. But a counterexample could lie beyond these ranges. That is precisely why they remain conjectures and do not become theorems.

This distinction is also relevant for AI. Language models, that is, programs that continue texts word by word, do not inherently know the difference between “proven” and “claimed.” They render both in the same confident tone. Anyone working with them must bring this distinction themselves.

Why language models invent such names

A language model predicts which word fits next. It has learned that after a proper name and a hyphen, “Conjecture” or “Theorem” often follows. It can apply this pattern without ever having seen the thing being referred to. The result sounds correct but means nothing. Experts call this hallucination: a fabricated piece of information delivered in the tone of a real one.

Rare proper names are especially susceptible to this. For the Goldbach Conjecture, there are thousands of texts a model has learned from. For a name that hardly occurs, this foundation is missing. The model then fills the gap with plausible-sounding text. A common misconception is that a model will say “I don’t know” in such cases. On its own, it rarely does.

One distinction helps: a search system either finds something or nothing. A language model always answers. That is why it does not replace research, but rather demands it.

What to do when a term appears nowhere

Such cases occur more often than one might think. Fabricated citations show up in term papers. News articles reference studies that do not exist. Even in court, lawyers have been caught citing freely invented rulings. The common denominator: someone adopted an AI-generated answer without checking it.

The practical approach is simple. One searches for the term verbatim in a search engine, on Wikipedia, and in specialized databases. For mathematics, arXiv or the Online Encyclopedia of Integer Sequences (OEIS) are suitable. If nothing turns up there, the term is very likely not an established one. In that case, it should not be used without openly flagging this.

If the “Fel Conjecture” originates from a specific book, paper, or course script, only the original source can help further. It will state who coined the name and what exactly is being claimed. Without this source, the term remains empty. Admitting this is more scientifically correct than supplying a plausible-sounding explanation instead.