Đang tải bảng tin…
Lobsters0 Hot0 bình luận53 phút đọc
How do programming languages impact token efficiency and correctness?
An empirical investigation into whether dynamic languages are more token-efficient and better suited for LLM coding agents than static languages. The author runs two non-trivial evals — implementing a zstd decoder from the RFC and a Pandoc-based TDD task — across many languages using GPT-5 via Codex. Results show that the widely-cited claim that dynamic languages are meaningfully more efficient or correct than static languages does not hold at higher effort levels. Language popularity shows a weak-to-moderate positive correlation with better outcomes. Obscure languages (J, Clojure on some tasks) and assembly perform poorly. The author also critiques methodological flaws in two existing public evals, including a critical bug where a Go agent symlinked a missing executable, causing all subsequent Rust tests to run the Go binary — invalidating the conclusion that Rust failed due to language difficulty. A third eval on implementing a board game (Guards of Atlantis 2) was abandoned as all agents scored near zero regardless of language.
Nguồn: https://danluu.com/pl-tokens. 8 Sync News chỉ tóm tắt và dẫn link; bản quyền nội dung thuộc tác giả và nguồn gốc.
