HN Showcase
Back to browse
Using LLMs and differential testing to convert code

Using LLMs and differential testing to convert code

by seddonm1·Feb 23, 2026·2 points·0 comments
Visit ProjectView on HN

AI Analysis

●●SolidBig BrainWizardry

LLM-powered decompilation with differential testing beats guesswork, but verification burden remains.

Strengths
  • •Clever use of Claude to handle borrow-checker and language idioms automatically, not just direct translation.
  • •Differential testing approach genuinely solves the hard problem: verifying correctness of LLM-generated code.
  • •Real-world validation on complex runtime code (spell casting logic) shows practical viability beyond toy examples.
Weaknesses
  • •Narrow audience: primarily academic or security researchers; most developers won't reverse engineer binaries.
  • •Verification still relies on behavioral testing—doesn't catch subtle memory safety bugs or side-channel issues in critical code.
Target Audience

Reverse engineers, binary analysis researchers, systems programmers

Similar To

Ghidra (NSA's reverse engineering framework) · Radare2 · Sourcegraph Cody for code migration

Post Description

This is the second part of a post aligned with a talk I gave recently at the Sydney Rust meetup.

I have found significant success in this process converting decompiled code and also converting source code from python to go. Give an LLM a target and they can be extremely stubborn (in a good way).

Similar Projects

AI/ML○Pass

Binary is no longer safe

Uses differential-property testing as an automated feedback loop to validate LLM-driven rewrites — that's the clever bit that turns flaky translations into repeatable refinement. The author targets a closed-source MUD DLL to avoid model memorization and walks through why raw assembly prompts failed and how decompiled C+tests + LLM translation to Rust succeeds. It's a thoughtful, slightly alarming demo with concrete techniques you can try yourself, not just vaporware.

WizardryBig BrainNiche Gem
seddonm1
303mo ago