+1 to this sentiment for now, I give them a try every 6 months or so to see how they advance. And for pure code generation, for my workflow, I don't find them very useful yet. For parsing large sets of documentation though, not bad. They haven't creeped their way into my usual research loop just yet, but I could see that becoming a thing.

I do hear some of my junior colleagues use them now and again, and gain some value there. And if llm's can help get people up to speed faster that'd be a good thing. Assuming we continue to make the effort to understand the output.

But yeah, agree, I raise my eyebrow from time to time, but I don't see anything jaw dropping yet. Right now they just feel like surrogate googler's.