Fascinating. I think static typed languages is the way to go regardless of token strength. The next step would be to see how many LLM tokens are used to build a project to a spec and testsuite using these languaes. I.e. realize end to end efficiency

Link to original post:

https://martinalderson.com/posts/which-programming-languages-are-most-token-efficient/

Sergey's headline is a little clickbait-y as the results put Haskell above F#. Also, the article itself admits that there are a lot of variables that were not accounted for in the test. It appearst to be a pattern of more concise languages performing better with LLMs than more verbose langauges with C being all the way at the bottom. I was under the impression that modern C# is more concise than Java, but I haven't touched Java in years.