Comment by lrvick
2 days ago
By the time something like this passes, everyone will have the ability to convert any game to unlicensed source code in a matter of hours.
Intellectual property is not going to be a thing anymore.
2 days ago
By the time something like this passes, everyone will have the ability to convert any game to unlicensed source code in a matter of hours.
Intellectual property is not going to be a thing anymore.
IP primarily existed for hundreds of years on things that are easy to copy. Technical difficulty is not relevant.
Hundreds of years ago, nothing was easy to copy, until the printing press. Early copyright surrounding that wasn’t about “IP protection” but rather who had the right to operate a printing press at all
And yet the clean room reverse engineering loophole remains in effect in the US.
It just used to take a lot of human labor, but LLMs are closing that gap. Fast.
You’re missing the point of the parent comment. IP exists BECAUSE things are easy to copy. If Apple REA’d Windows and sold it, Microsoft will get paid BECAUSE of IP. The system is working as designed.
3 replies →
You're dreaming. Prepare for DMCA2, where it will be illegal to run an LLM on a compiled binary for the purposes of decompiling under penalty of jail time.
As long as it is legal in at least one country, the code will be out there and models will train on it. The ship on IP has already sailed. The moat is gone.
Which models? Of Iranian Army or what? (this was sarcasm)
AI companies tend to discriminate against a lot of data they gather when it comes to training phases, usually for legal reason. Do you have any concrete proof that game decompilation efforts influenced any "mainstream" models?
1 reply →
With one giant exception, the advent of the algorithm all over again but with new proprietary and novel algorithms sets.
How is that working out for the mp3 algorithm or bluray decryption?
The OSS community just laughs at algo patents. There was never any ability to enforce them at the pre-AI scale. Post-AI the corpos have no chance. They lost.
Notice the word new, novel and proprietary. Would you like to ask your AI pipeline to duplicate ours? Unless someone knows what we do to parameterize, create and optimize new, proprietary and novel algorithm sets, it remains as part of any new scientific or engineering moat associated to an AI pipeline. This also include deep knowledge domain expertise and human decision trees applied along the way.
2 replies →
If I could run SoTA models on my desktop, you would be onto something. While it's a subscription, the big guys who own the data centers will do whatever the guys in government tell them to. TLDR: you'll start getting the "I can't assist you with this task" reply to all your prompts.
> everyone
Assuming everyone is provided unlimited compute in the future is very naive
I already have built multiple confidential compute operating systems from 386 bytes of machine code all the way up with many novel and world-first features using exclusively consumer hardware I personally own, without any internet reliance, on my own property. Low level kernel and compiler debugging all with models that will run on old GPUs you can buy on ebay right now for under $500.
The way most people use AI, lighting money on fire, is a skill issue. You can build anything cheaply now if you generally know how computers work at a low level and can tell models exactly what to do.
Link to cheap processors and storage?
[dead]
That's why they're pricing out desktop builders and trying to lock everyone into hosted model consumption from a cheap mobile.
And if you can afford it a VR headset or AR glasses to simulate an 80" display