Breaking news, the Collatz conjecture was false. Verified false, like, formally verified false using Lean for 2.5 days in July. The proof was approved in Lean for two and a half days until they fixed the bug in the kernel in Lean that this proof was exploiting, which of course was not a proof. It was just an exploit of a bug. And the crazy thing is that this proof of the Collatz conjecture was also verified using the comparator, which is supposed to be a trustworthy judge for Lean proofs, making sure that they match the theorem they are supposed to be proving, that they are not using some external axioms that are just wacky or just the axioms assume what you are trying to prove, etcetera. So let me explain what happened. On July 25th, a user on GitHub, Ramana Kumar, posted a disproof of the Collatz conjecture. This was a package in Lean where he first formalizes the Collatz problem, the Collatz conjecture, and then he introduces a formalized theorem and proof of the falseness of the Collatz conjecture, where the theorem shows that there is some orbit that does not come back to one. And the package includes a verification using Lean and Mathlib code that, yes, that this entire package is correct. Quickly, the bug in the Lean kernel that was causing this problem was identified and by July 28th, early at 1:00 a.m. in July 28th, a fix was submitted to patch this bug. Apparently, the bug was with how nested inductive types were coded into Lean and that with, like, ill parameters, they would still be accepted by the kernel, and this was, or it has been fixed now. And people are rightfully concerned about this bug because it has gone unnoticed in lots of stable versions of Lean 4. In fact, it seems that the way this bug, or this, like, fake proof was found was because they, perhaps Ramana Kumar or some other user, they were using LLMs, AI, to look for vulnerabilities in Lean that could be exploited somehow, and they found this, sort of, inductive types problem, and the Collatz conjecture was a natural kind of problem that one could use to showcase how devastating such a bug could be to Lean. And there has been some amount of well-founded panic in the Lean community, and people asking, how is this possible? How did this happen? How did such a bug go unnoticed for so long? And are there other bugs, other vulnerabilities in the Lean kernel that are yet to be discovered, and they could be producing errors in the logic of Lean proofs? I just want to finish this video by saying that the Lean community has all my respect. This is an incredibly important project, especially in this era of math and AI, and it is built by a community of volunteers. So there are going to be steps like this where there is some serious issue, bug, vulnerability that is found in the Lean code, so they might have to take some steps back to make sure that they are implementing more security checks along the way, but overall they are doing an amazing work.