Post

seems conspicuous that programmer commentary on genAI is split between "oh yeah i use claude all the time now. the 10× programmer is real and he's me. i am a god" and "yeah my coworkers use claude all the time. they keep committing crap that doesn't work and now i spend half my time fixing it"

One of my coworkers wrote some Python code in a weird, roundabout way. I asked him why he didn't do it the obvious way. "ChatGPT said it wouldn't work." The obvious way works in the Python 3 version we're using. It works in Python 2.7. Apparently, "Just try it and see what happens." is a lost art.

Not only did I have to try it for him, but I ended up distracted by the question of how the hell this misinformation got synthesized, because I couldn't find any instance of anyone or anything making such a claim by just searching the web.

The task in question was "define a property on a NamedTuple". Somehow the Statistically Average Coder had generated the assertion that the property has to be patched onto the class after it's created.

Starting in 3.6, typing.NamedTuple is doing some kind of metaprogramming stuff so you can subclass it directly and use the attribute annotation feature to name the tuple fields. I just looked at the implementation, and, um, if it works...

I thought I was coming up with out-there metaprogramming techniques in my own projects, but github.com/python/cpyth... shows me that I am like little baby.