Who is using AI to code? Global diffusion and impact of generative AI

The "global diffusion" is diffusion of generative "AI" itself, not basic knowledge. Since "A" scrapers have basically ingested the entire web (including Wikipedia and Reddit and Stackexchange), an LLM probably does contain most or all human knowledge, at least syntactically.

Someone asking, say, ChatGPT for some info might actually get a true answer that they'd never heard of before. In that way, "AI" might help spread or "diffuse" knowledge all over the place rapidly.

An example: I reviewed an "AI" generated pull request at work. It included a large-ish bash script. The "AI" was unable to maintain a single style throughout using if [ $X = "some string" ], if [[ $x == "some string" ]] and if [[ $X = *"some string"* ]] with abandon. Notably, it did not use the bash regular expression matching operator, =~.

I was unfamiliar with getting bash to do matches by globbing with if [[ $X = *"some string"* ]] . I learned something from an "AI" that day.

Instances like this diffuse knowledge via "AI". I want to know if "AI" helps diffuse knowledge, not if it promotes productivity or some other pointy-haired-boss thing. Using lines of "AI" generated Python code as a measure of productivity is something only director-level managers could love.