I wanted to calculate the number of words added between commits in my blog posts. I asked Gemini for a Nushell pipeline based on git diff:

It proposed the following:

git diff --word-diff=plain -- "*.md" | rg -o '\{\+(.*?)\+\}' | str replace -r '\{\+|\+\}' '' | str join " " | str words | length

However, I ended up using this simplified version:

git diff --word-diff=plain -- "*.md" | rg -o '\{\+(.*?)\+\}' | str replace -a "{+" "" | str replace -a "+}" "" | str replace -a ' ' "\n" | lines | length

I’ve noticed that LLMs often lean toward overly complex solutions. I’m not sure if this is due to being trained to generate as many tokens as possible, but using regular expressions where a simple string replacement would suffice is often a sign of unnecessary complexity.