linuss-law ADG-0018

Linus's law

“Given enough eyeballs, all bugs are shallow.”

Origin

Eric S. Raymond, 1997, "The Cathedral and the Bazaar", named for Linus Torvalds. About open-source review: many independent reviewers with different backgrounds find what one misses.

Why it should, or should not, apply to models

The word doing the work is "enough", and it hides "independent". Human eyeballs differ in what they know and what they miss; many instances of one model share one set of blind spots, so adding them adds count without adding coverage. The law should transfer only when the extra eyes are genuinely different — a different model, different context, or a non-model sensor — and should fail outright when the extra eyes are the producer looking again.

Standing: Holds, narrowed

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

Where it breaks

Claims showing the heuristic failing for models.

In multi-agent LLM systems where one agent starts with an erroneous shared belief, broadcasting every agent's full context at every step raises the rate of false assertions above doing no synchronization at all, because the error propagates to agents that were previously correct — and the harm appears only in tasks where one wrong fact cascades across semantically linked dimensions (destination to airport to weather), not where agent contexts are largely orthogonal.
Stating false facts confidentlyReviewed by AI
More agents with shared context made a false belief more common, not less. Eyeballs that share a context are one eyeball.

Where it holds only under a condition

Claims that keep the heuristic but bound it.

Without an external, ground-truth signal — a failing test, a compiler error, a verifier's output — a model's own critique of its reasoning is not a reliable improvement signal, and asking it to review and revise a correct answer often turns it into a wrong one.
Fixing its own mistakesReviewed by AI and Russ Miller
The producer looking again is not another eyeball. Without an external signal, self-review does not make the bug shallow and can introduce one.

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.

Notes

Wanted: a measured comparison of review by N instances of one model against review by N different models on the same defect set. Ensemble and mixture-of-agents papers should carry this.