On troubleshooting, either LLMs used to be better, or I'm in a huge bad luck strake. All of the last few times I tried to ask one, I've got a perfectly believable and completely wrong answer that weren't even on the right subject.
On code review, the amount of false positives is absolutely overwhelming. And I see no reason for that to improve.
But yes, LLMs can probably help on those lines.
