brandolinis-law ADG-0006

Brandolini's law

“The amount of energy needed to refute nonsense is an order of magnitude bigger than that needed to produce it.”

Also: Bullshit asymmetry principle

Origin

Alberto Brandolini, 2013, a tweet about Italian political debate. About the asymmetry between asserting and checking.

Why it should, or should not, apply to models

A language model is the cheapest generator of fluent, plausible text ever built, and fluency is exactly what makes a false statement expensive to check. So the asymmetry should widen: generation cost falls toward zero while verification still needs retrieval, a test, or a person. The law transfers on both sides of the model — it is expensive to check what a model produces, and a model asked to check something is itself a cheap producer that struggles to do the expensive refuting.

Standing: Holds

Standing counts claims a person has reviewed. Claims reviewed by AI alone are listed and marked, and do not move it.

Where it holds

Claims showing the heuristic applying to models.

Models endorse widely held falsehoods, showing weak verification against what they know.
Checking claims against evidenceReviewed by AI and Russ Miller
Asked to refute, the model endorses instead: producing agreement is cheap, verifying against what it knows is the expensive step it skips.
Surveys causes, kinds, and detection of hallucination across data, training, and inference, and finds no complete solution.
Stating false facts confidentlyReviewed by AI and Russ Miller
Detection is the unsolved side. The survey finds no complete method for catching what is trivially cheap to generate.
General-purpose factuality verifiers (entailment-, consistency-, and RAG-trained) and zero-shot frontier LLM judges perform near chance at spotting unsupported claims in scientific peer reviews that must be checked against the full submitted paper, because they cannot separate paper-unsupported assertions from legitimate evaluative critique; small models fine-tuned on in-domain examples beat them by a wide margin.
Stating false facts confidentlyReviewed by AI
Dedicated verifiers and frontier judges perform near chance at spotting unsupported claims — the refuting side is where capability runs out.

Add a case where it holds or breaks

A break is worth more than a hold. If you have seen this fail for a model, with the setup written down, that is the most useful thing you can add here.