logoalt Hacker News

thejokeisonme • today at 4:05 PM • 2 replies • view on HN

A fake refutation of the Collatz conjecture already got through one of these independent checkers in the past.

What's harder, proving GRH or finding a way to get through every checker (which aren't as independent as they are sometimes touted to be)?

In this case there was a human expert in the loop so it's much less likely the model hacked its way through the problem without the expert noticing. But the statement as written has 4 points, none of which rests on the human involvement. So I don't like that they claim full confidence based on those points.


Replies

TwoFx • today at 4:13 PM

A lot has happened since the fake Collatz refutation: https://leodemoura.github.io/blog/2026-10-15-lean-now-ships-...

In short, Lean has formally verified checkers now, and the RH proof was checked with them. So any such soundness hole would have to be exploited in the correctness proofs of the checkers, which is a much smaller attack surface than the RH formalization.

➕ show 2 replies
dev_dan_2 • today at 7:33 PM

> But the statement as written has 4 points, none of which rests on the human involvement.

The theory and implementation of Lean is very much so a result of human involvement, on a massive scale even. Just because it is something repeatable does not lower that in any way.

Distrust of LLM output should not turn into distrust of automation, I think. The former I totally get (and deeply hold myself), the later is likely to hold one back - imagine banning calculators because there is not human in the loop, to make an extreme example.