Hacker News Reader

Comment by jameshart | Hacker News Reader

This feels very much like the work of someone with ‘just enough knowledge to be dangerous’.

At no point in this process does the author seem to stop and inspect the results to see if they actually amount to what he’s asking for. Claiming that this output represents a decompilation of the obfuscated target seems to require at least demonstrating that the resulting code produces an artifact that does the same thing.

Further, the claim that “Using the above technique you can clean-room any software in existence in hours or less.” is horrifyingly naive. This would in no way be considered a ‘clean room’ implementation of the supplied artifact. It’s explicitly a derived work based on detailed study of the published, copyrighted artifact.

Please step away from the LLM before you hurt someone.

View on HN

“Using the above technique you can clean-room any software in existence in hours or less.”

Having spent my misguided youth doing horrible things to Sentinel Rainbow and its cousins - I can only chuckle.

For those who don’t know Sentinel Rainbow is a DRM dongle these were popular in the 90’s through early 2000’s for enterprise/business software especially that could be useful on the smaller end of the businesses scale (CRM, ERP, CAD/CAM) where piracy concerns were much bigger due to the relative high cost of development combined with a relatively small market to begin with.

SAAS pretty much made all of that obsolete, since with SAAS you get an unbeatable DRM for free.

If my understanding is correct, what's legally protected is reproducing a proprietary IP design rather than studying the design?

If you create a new design that doesn't have the proprietary elements in it that's not grounds for copyright infringement?

If you studied the source code, you can't say you've independently created the copy, the trick with clean room design is the implementers are going off a specification which does not include copyrighted material.

https://en.m.wikipedia.org/wiki/Clean-room_design

But maybe you could use one LLM to study the software and write a specification, then throw that over the wall to a different human who uses an LLM to write software based on that spec

So you should do the same as with normal clean-room practices?

LLM1: code -> english description of code

LLM2: english description of code -> code

And that would be clean room? Might be cool to automate that. I bet you could train LLMs to do exactly that.

I checked the author’s result. It is 100 percent BS. All hallucination. Has nothing to do with the original claude code.

Reply on news.ycombinator.com