Basis uses LLMs and Lean to verify an interpreter down to RISC-V machine code
burny_tech · x · 2026-10-26
Basis published "Ship Your Interpreter": LLMs wrote a formal semantics for a small programming language, then Lean proved the interpreter follows those rules even after compilation to RISC-V machine code — proving a refinement against your existing codebase's binary, so "slop in, trust out."
The article traces recent progress: in April their agents built a JS-to-WebAssembly compiler but couldn't prove behavior preservation; later they verified Linux's nftables compiler; in September MIT's Kaashoek and Zeldovich verified the xv6 kernel down to machine code using AI agents.
Basis argues the bottleneck is shifting from theorem proving to specification: ensuring what we prove captures the behavior we actually care about. Their goal is verification methods that scale to all software, making critical infrastructure secure by design.
More from coding & agent
- Base44 fine-tuned Qwen 2B in half a day for prompt autocomplete serving millions — tristanbob · 2026-10-26
- Grok highlights Blender MCP demos: FIFA-style playable game from a single prompt — chongdashu · 2026-10-26
- Grok Bot template lets ChatGPT and Claude Code hand off tasks via webhook — xiaohu · 2026-10-26
- Zero coding background, real app built with ChatGPT: a user's lightbulb AI moment — wishfulthink95 · 2026-10-26
- Nearly 60% of AI-written formal specs proved the wrong thing, study finds — burny_tech · 2026-10-26
- Jalapeño chip is just a systolic array with HBM: why custom inference silicon is easier than you think — ai · 2026-10-26