Well, the tool apparently fought him and “several times stated flat out that this was impossible and unsolvable and that we should just write a report about it.”

And it also doesn’t look like the LLM actually added code so much as helped with debug code and analysis, however exactly that may have looked like. The resulting fix changes three lines of code, very much human-writable or -verifiable.

Also, let’s not worship Linus. The guy has done much good and is brilliant, but he’s still human, still capable of making mistakes. This instance working out well deserves the proper appreciation, bu we still ought to be wary.