Rust Project Perspectives on AI: Practical Insights from Contributors

What This Is

A summary document authored by nikomatsakis on Feb 27, 2024, collecting perspectives from Rust contributors and maintainers about AI tool usage. The document aims to capture the full range of opinions to understand the landscape of arguments on each side.

Key Details from the Source

The Rust project does not currently have a coherent view or position around AI tool usage. This document represents individual perspectives, not an official project stance. The discussion covers both AI use on rust-lang crates and AI use by Rust developers elsewhere.

AI Requires Engineering Skill

Those getting the best results from AI emphasize it takes real engineering to make AI work well. As contributor TC notes: "It takes care and careful engineering to produce good results. One must work to keep the models within the flight envelope. One has to carefully structure the problem, provide the right context and guidance, and give appropriate tools and a good environment."

TC also observes rapid improvement: "Something that might not be obvious is how much things have changed over the last 2-3 months. At one time, it was hard to justify the use of models for serious work. But the state-of-the-art models are now too good to ignore."

Non-Coding AI Use Cases

Many contributors find value in AI for tasks beyond coding:

• Searching and discovery: davidtwco reports using internal AI tooling at Arm to search through 10,000+ pages of architecture documentation, making it easier to respond to upstream issues promptly.
• Codebase navigation: scottmcm finds AIs helpful for research questions like "well I'm here and I need a Span; where do I get one?"
• Code review and idea exploration: BennoLossin uses AI for double-checking work and asking questions that help explore correct ideas. RalfJung mentions interest in exploring LLMs for code review, citing Linux kernel folks having success with project-specific, carefully crafted prompts.
• Large-scale data processing: Several contributors note AI makes working with semi-structured data more tractable, with examples from the FLS (Future Language Specification) group.

Differing Experiences

Contributor yaahc notes the cognitive dissonance between respected developers finding value in AI tools while others find "99% of the value people claim from these tools to be all smoke and no substance." The difference appears to stem from how inputs are structured and how tools are used, with experienced engineers achieving better outcomes.