AI formal verification hits reality: the halting problem blocks existing codebases
danbri · x · 2026-10-06
A retweet of a long thread by @hillelogram dissecting a common misconception about AI doing formal verification.
People assume that in an AI-does-formal-verification world, you can hand your existing codebase to AI, ask it to prove correctness, and get either a proof or a bug. Reality has bones to pick — starting with the "existing codebase" part, which runs into the good old halting problem: no general-purpose algorithm can determine, for any program and input, whether it terminates.
Extended via Rice's theorem, no nontrivial semantic property of programs can be automatically proven for arbitrary programs. We can still prove some things about some programs — that's why formal verification exists — but the trick is writing code in a style suitable for verification, suggesting AI verification at scale requires changing how code is written, not just bolting it onto legacy codebases.
More from coding & agent
- invideo launches MCP with 250+ SOTA models for full-stack AI video production — aziz4ai · 2026-10-06
- Same models, new harness: scores jump 23% to 62%, says Rajiv Shah — rajistics · 2026-10-06
- Harvey's rebuilt harness lifts 7-model average from 23.3% to 62.4% while cutting cost 60% — rajistics · 2026-10-06
- Dell expands AI Data Platform with semantic layer and knowledge graph to make data agent-ready — DavidLinthicum · 2026-10-06
- llmClassificR update brings full LLM text classification pipeline to pure R — GolinoHudson · 2026-10-06
- Small businesses quoted $30K for AI agents an experienced builder ships in weeks — Warm-Reaction-456 · 2026-10-06