[applfanboysbgon]

As someone who reverse engineers games for fun and preservation, I would personally be more than happy to take an external request. With compensation, of course. Maybe not so much if it's the "do hundreds to thousands of hours of highly specialised labour for free" kind of request :)

[thaumasiotes]

I transcribed several pieces from a public-domain orchestral score of Swan Lake and uploaded them to Musescore's web site. This would have been ~dozens of hours of work. (For me, starting with almost no relevant training.)

I did this because those particular pieces were significant to me.

But I attracted the attention of another user who wanted to see more public-domain orchestral music on the site, and who contacted me to ask if I was planning to do the entire ballet (no; it's about 600 pages) and if I would take requests.

I responded that I was happy to take requests but wouldn't guarantee that I'd do any work on them. He requested Tchaikovsky's Italian Capriccio, which was a decent guess as to what I might be willing to transcribe (it's the same author)... but I looked into it, listened to the music, and just couldn't muster the enthusiasm to keep working on it.

So yeah, a request might work, but it's only likely to work if you happen to make an excellent guess about what the person would have wanted to do anyway. Think of it like drawing someone's attention to something.

[qingcharles]

What is the current SOTA in terms of tools? How are these tools being used with LLMs to speed up decompilation?

My current wishlist is to decomp Elite CGA version (tiny x86 binary) back into assembler and annotating all the method names, vars etc. That way I could swap out some of the inner loop using knowledge that has been uncovered in the last 40 years of optimizations.

[flakes]

These models are getting crazy good at examining things like core dumps and disassembly. I've been using an agent to write compiler logic, and its amazing the kind results you can get by having the agent examine the raw binary outputs. I would not be surprised to see agents excel at identifying and labeling patterns for decompilation.

[gspr]

Really? Could you share your techniques that get you there?

Inspired by https://github.com/scosman/cursed_browser, I have a little art project going where the CPU of a virtual machine is entirely LLM-powered. But even though the ISA is well known and clearly in the LLM's training data (it answers question about it mostly fine), I can rarely get it to even decode a handful of instructions in a row correctly. It'll e.g. do 10 instructions right (even execute right!), then just lose the ability to do bit manipulation all of a sudden and fail miserably at even decoding the 11th. If I try to help it along it'll apologize profusely, do it wrong in five novel ways, before it gaslights me saying I'm in fact mistaken.