Decompiling 2024: A Year of Resurgance in Decompilation Research

Re:Decompiling 2024: A Year of Resurgance in Decompilation R... by sitkack on Thursday January 30, 2025 @11:22AM (#42879258)

You are right on a lot of things, but LLMs are the best bijective lens that humanity has ever discovered. They can invert functions we didn't think were invertible.

If given a mostly correct transform from binary back to code, how would we fix that?

Exactly!

Heuristics are dead.

[ Parent | on HN ]

Re:Decompiling 2024: A Year of Resurgance in Decompilation R... by thesz on Thursday January 30, 2025 @02:08PM (#42881011)

> Heuristics are dead

I guess it is an example of an heuristic.

[ Parent | on HN ]

Re:Decompiling 2024: A Year of Resurgance in Decompilation R... by ahartmetz on Friday January 31, 2025 @12:33AM (#42884964)

Absolutely. LLMs use methods that are not really provably solving the given problem, they just happen to work in most relevant cases. That's a heuristic.

[ Parent | on HN ]

Re:Decompiling 2024: A Year of Resurgance in Decompilation R... by sitkack on Friday January 31, 2025 @08:00AM (#42887220)

http://www.incompleteideas.net/IncIdeas/BitterLesson.html

> Our methodology leverages state-of-the-art natural language processing techniques to systematically evaluate the evolution of research approaches in computer vision. The results reveal significant trends in the adoption of general-purpose learning algorithms and the utilization of increased computational resources. We discuss the implications of these findings for the future direction of computer vision research and its potential impact on broader artificial intelligence development.

https://arxiv.org/abs/2410.09649v1

[ Parent | on HN ]

Re:Decompiling 2024: A Year of Resurgance in Decompilation R... by ahartmetz on Friday January 31, 2025 @08:47AM (#42887592)

I know that these algorithms are the best that exists for many tasks, but any non-deterministic non-provable algorithm is still technically a heuristic. Also, currently, the bitter lesson seems to be that more of the same runs into diminishing returns, contrary to "The Bitter Lesson"(tm). There will probably be a better AI architecture at some point, but "lol just add more GPUs / data / RAM" times are kind of over for now.

[ Parent | on HN ]

Re:Decompiling 2024: A Year of Resurgance in Decompilation R... by sitkack on Friday January 31, 2025 @11:12AM (#42888932)

You are misinterpreting the bitter lesson, it only gets more bitter with time!

The arxiv paper is well written, worth a read.

[ Parent | on HN ]

Re:Decompiling 2024: A Year of Resurgance in Decompilation R... by ahartmetz on Friday January 31, 2025 @12:48PM (#42889784)

I don't think that I misunderstood - it says that more computing power beats "hand-engineering". But more computing power doesn't seem to work that well anymore to improve AI performance - and there is not much more computing power coming in the next couple of years, not orders of magnitude more for free like it used to be. Case in point: the somewhat disappointing new nVidia generation.

[ Parent | on HN ]

Re:Decompiling 2024: A Year of Resurgance in Decompilation R... by sitkack on Friday January 31, 2025 @03:16PM (#42891414)

> it says that more computing power beats "hand-engineering".

This is the simplistic tl;dr, but that isn't what is says. You should reread the essay and the linked arxiv paper.

It is much more than throwing more compute at a problem.

[ Parent | on HN ]