Русский(RU) Я дал LLM писать unsafe Rust полгода. Miri плакал Полгода я давал LLM писать unsafe Rust в боевых проектах и разбирал каждый блок под miri и санитайзерами. Кате

LLMs consistently produce unsafe Rust code errors

作者 PulseAugur 编辑部 · [1 个来源] · 2026-05-19 11:12

A developer spent six months having large language models write unsafe Rust code for production projects. The models consistently made specific types of errors, including issues with aliasing, provenance, manual memory management, and concurrency in FFI callbacks. Each category of error was documented with minimal examples and provided fixes. AI

影响 Investigating LLM code generation quality reveals persistent safety vulnerabilities in complex programming languages.

排序理由 The cluster describes an experiment and analysis of LLM-generated code, fitting the research category. [lever_c_demoted from research: ic=1 ai=1.0]

在 Mastodon — fosstodon.org 阅读 →

AI 生成摘要 · Google Gemini · 来自 1 个来源。我们如何撰写摘要 →

报道来源 [1]

Mastodon — fosstodon.org TIER_1 Русский(RU) · [email protected] · 2026-05-19 11:12

I let LLMs write unsafe Rust for six months. Miri cried. For six months, I let LLMs write unsafe Rust in production projects and analyzed every block under Miri and sanitizers. Kate

Я дал LLM писать unsafe Rust полгода. Miri плакал Полгода я давал LLM писать unsafe Rust в боевых проектах и разбирал каждый блок под miri и санитайзерами. Категории ошибок, которые модели делают стабильно: aliasing, провенанс, layout в alloc/dealloc, забытый ManuallyDrop, гонки …

链接 habr.com/…/1035914

报道来源 [1]

I let LLMs write unsafe Rust for six months. Miri cried. For six months, I let LLMs write unsafe Rust in production projects and analyzed every block under Miri and sanitizers. Kate

相关实体

相关话题