Stolen assets with no compensation being rendered by code built by no one.
We had piracy before LLMs. It was more honest.
Another example recently is Ghost Recon 2; which originally came out on Xbox has now been ported to the PC.
https://www.ghostrecon.net/forums/index.php?/topic/75174-gho...
I never got a chance to play it as I was mainly a PC gamer, so is that a bad thing to now have that opportunity to play it 23 years later on my PC?
The idea of reimplementing the _engine_ is good. Handing out the _assets_ however is not.
The trajectory is clear I'm afraid. Games will in the future be streamed from the cloud, so that there are neither assets nor logic to steal.
Previously, before LLMs, the people doing this stuff had enough friction to consider their actions. Now they don't.
Stadia is already dead. Do you think someone else will pick that up?
To be honest I don't see the connection to LLMs with that particular problem (original assets for download). This could have been done without LLMs in the past. It was always illegal, it is now illegal and therefore it will be prosecuted if there's someone interested in it.
That is a good point but could it be also for the following reason; cloud streaming will be "encouraged" as hardware prices keep high and will encourage a monopoly on data centres to protect those hardware asset values?
But just because there already is fuel, doesn't mean that one has to add even more to the fire
Cloud can be beaten too - once you can decompile engine, what stops you from simply fetching all the assets with 1 legal copy from cloud, and repackage game with all of them, with no internet required and piracy checks removed? Its a race and no side is considering dropping their approach.
I think they meant the kind where you only get a video stream, can't do anything there.
Also aren't the AI models themselves copyright violations? On a much larger scale than anything they produce? It is now proven (they admitted in court, so this can be submitted in other cases as evidence and can't be disputed) that OpenAI, Anthropic and Facebook/Meta all trained their models on illegally downloaded copyrighted material, and they did so without a license to even read that. They also did that before the AI training copyright exception came into force, so even that doesn't cover it.
How reasonable is it for courts to allow the AI companies to do that, then have problems with what is produced with the models?