I don't think I can notice bugs by reading code. I think I can notice bugs by:

* Writing code in a language with a static type system.

* Reading the code and "playing computer".

And, last but by far the most important:

* Actually testing the code

Turns out that using an LLM to generate code doesn't preclude testing it!

Writing tests after the fact is way harder than writing code to be testable. You're pretty much guaranteed to miss edge conditions, which tend to be the cases LLMs often screw up.

The rampant optimism is just weird to me.

You seem to be assuming that we're generating large amounts of code with LLMs that we don't read and adapt and make our own. I can definitely see that as a risk for overconfident juniors, but that is not the MO for senior developers who are using these tools.

95% of the code that I add from an LLM is less than a line, a nice autocomplete. The other 5% is a function that I have ChatGPT get started for me and then I read it and refactor it until I understand it as well as any code I've written myself.

The rampant fearmongering is just weird to me. It's a tool, and like any tool some people abuse it, but it's weird that some people simply can't believe that there are people who use it well.

I see people overestimating their ability to notice bugs.