Speaking as a software engineer, I'm not at all threatened by it. I like Copilot as fancy autocomplete when I'm bashing out code, but that's the easy part of my job. The hard part is understanding problems and deciding what to build, and LLMs can't do that and will never be able to do that.

What I am annoyed by is having to tell users and management "no, LLMs can't do that" over and over and over and over and over. There's so much overhype and just flat out lying about capabilities and people buy into it and want to give decision making power to the statistics model that's only right by accident. Which: No.

It's a fun toy to play with and it has some limited uses, but fundamentally it's basically another blockchain: a solution in search of a problem. The set of real world problems where you want a lot of human-like writing but don't need it to be accurate is basically just "autocomplete" and "spam".