Comment by datsci_est_2015
9 hours ago
I also find it impossible to parse half the comments that Claude tries to sneak into our pull requests. I’ve never had an issue understanding code comments written by humans like this before. The structure of the information is like a waterfall that leaves me unable to swim to the surface and grab the air of comprehension.
So, “Please write a one-liner comment manually to replace these 5 lines of AI generated comment” is a common refrain in my PR reviews to colleagues.
We're using grok and while I can usually understand the comments they're always at least 2x as verbose as they need to be. It loves explaining everything in two different ways, putting one of the explanations in parenthesis.
It is getting to the point where you are better off getting an LLM to describe the PR changes in your preferred style (make it very short, make a table describing API changes, list renames in a table, etc, etc) than go to the one created by the reviewer. If only the reviewer could embed his prompts (like "make clear X" is happening) or his manual edits.
But to be honest I doubt most people who use AI for PR descriptions even bother changing anything.
Usually, if you don't understand something in what an AI writes, it is a clear sign that there is a problem hidden somewhere in there. I explicitly always ask wtf exactly it means by something I don't understand, and for sure there is a problem there. AI is pretty good at isolating a problem, giving it a cute name, declaring it solved modulo cute name, and moving on.
Personally I don't even read the PR description anymore but just the code, it's easier to understand what the AI is doing by reading the code rather than the word soup it tried to make