I hate that the author is correct: AI-generated bug reports have that weird jargon about “gated” array access, every version or test is “pinned”, every test case is a “fixture”. Between the omnijargon and the meandering sentences, I’m always about half certain that any given bug report is nonsense. But this article does explain the tension between real, albeit exotic issues and figuring out inhuman verbiage.

score 1 · canonical

OK: “keep a map of variable names to meanings in their context”.

score 1 · canonical

Does “differently difficult” entail substantially different “Best Practices” and paradigms?

Basically all of the programming things, from variable names to functions to object orientation exist so that humans, who at best have limited cognitive resources (as this article notes!) to write working, more-or-less correct programs. If LLMs take over, do easy things like “Best Practices” change? Are global variables acceptable if the LLM has an unlimited working memory instead of 5 +/- 2 items? Is “object orientation” necessary for something that doesn’t have cognitive biases and respects data structure boundaries? Will memory leaks become a thing of the past when the programmer LLM never forgets? Even very simple things like descriptive variable names don’t matter if the LLM can keep a mental map of variable name to meaning. I suspect iterative development like Agile will still work best, because the requirements (a.k.a. “prompting”) are never complete or finished, or even accurate.

I’m really curious to see how this category of thing turns out.

score 1 · canonical

I’m amused that Aho-Corasick structures can be constructed easily from trees, because Aho-Corasick can be modified to do glob-style matching of trees.

It looks like all the nifty transition diagrams were done using GraphViz. GraphViz is very useful.

score 1 · canonical