I think copyright law has simply failed to keep up with the reality of technological progress.
I don’t really see a problem with LLMs slurping up and regurgitating the world’s knowledge, if it was all freely and equitably available, and more importantly had been beforehand. We’d all herald it as an incredible innovation. And in some ways it is even now, because I do have access to it in numerous forms.
The problem is that some author’s livelihood is adversely impacted by this, as copyright is (in a sense) what ensured their remuneration for the original work. So LLMs undermine the existing social and financial order in deleterious ways. It’s the same thing we’ve seen play out in mathematics, a bunch if tech bros show up with a powerful tool, use it in a brash manner that likely causes real harm, then worse, start pulling up the ladder when big media conglomerates complain.
I suspect we’ll soon be in a situation where we no longer have easy access to unfiltered LLM outputs. If your OSS project was in the training corpus it will still retrieve it. It will not, however, retrieve content from NYT or other protected sources. The end result is that small-scale, solo creators are damaged, and large-scale creators are still afforded copyright protection.
There is no way to enforce opennness. The LLM gives me output, I iterate on it by hand and with other language models, who does it belong to now? We are ushering in a post scarcity world for a select few.
I don’t really see a problem with LLMs slurping up and regurgitating the world’s knowledge, if it was all freely and equitably available, and more importantly had been beforehand. We’d all herald it as an incredible innovation. And in some ways it is even now, because I do have access to it in numerous forms.
The problem is that some author’s livelihood is adversely impacted by this, as copyright is (in a sense) what ensured their remuneration for the original work. So LLMs undermine the existing social and financial order in deleterious ways. It’s the same thing we’ve seen play out in mathematics, a bunch if tech bros show up with a powerful tool, use it in a brash manner that likely causes real harm, then worse, start pulling up the ladder when big media conglomerates complain.
I suspect we’ll soon be in a situation where we no longer have easy access to unfiltered LLM outputs. If your OSS project was in the training corpus it will still retrieve it. It will not, however, retrieve content from NYT or other protected sources. The end result is that small-scale, solo creators are damaged, and large-scale creators are still afforded copyright protection.
There is no way to enforce opennness. The LLM gives me output, I iterate on it by hand and with other language models, who does it belong to now? We are ushering in a post scarcity world for a select few.