Hacker News Reader

Comment by lubesGordi | Hacker News Reader

I was just skimming this and I thought there were easy ways of decompiling into C source as well with generated function names like a, b, c or whatever. Could one just dump that sort of minimized C code into an llm and have it generate variable and function names maybe? I'd be curious to see if that worked.

View on HN

IDA and Ghidra, which they mention at the bottom, along with Binary Ninja which they didn't, have decompilers that will output C source-ish from compiled assembly. Variables and functions without symbols in the binary are just given unique non-meaningful names. There's a lot of AI-powered extensions for them now trying to automate the reverse engineering process of giving them meaningful names, but IMO they are completely worthless and of negative utility - they're either so high level they just reiterate what the assembly is doing but in English, or are wrong and make up explanations for nearly every name that are incorrect and take more effort to correct than they save.

I found this[0].

[0]: https://thejunkland.com/blog/using-llms-to-reverse-javascrip...

Some software might even get more readable by using minification and this tool afterward :-D

At least it would bring some normalization to variable and function names.

I had the same thought. Couple it with a source fuzzer to generate a large training dataset of valid source, and you could make a very nice disassembler.

Reply on news.ycombinator.com