this post was submitted on 03 Sep 2024
1574 points (97.8% liked)
Technology
59454 readers
4342 users here now
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related content.
- Be excellent to each another!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, to ask if your bot can be added please contact us.
- Check for duplicates before posting, duplicates may be removed
Approved Bots
founded 1 year ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
That's not a meaningful distinction, I spent all day using a Copilot search engine because the answers I wanted were scattered across a bunch of different documentation sites.
It was both using the AI models to interpret my commands (not generation at all), and then only publishes content to me specifically.
I’m talking about the training phase of LLMs.that is the portion that is doing the scraping and generation of copy written data.
You using an already trained LLM to do some searches is not the same thing.
Technically it is meaningful, fair use is for specifically things that don't replace the original in function.
Depends on what the function was. If the function was to drive ad revenue to your site, then sure, if the function was to get information into the public, then it's not replacing the function so much as altering and updating it.
If that "altering and updating" means people don't need to read the original anymore, then it's not fair use.
TBH I'm for reigning in copyright substantially, and would be on the shitty text generator company side of this, but only if it makes a precedent and erodes copyright as a whole instead of just creating a carveout if you have a lot of moeny for lawyers.
I generally agree, but I really think people in this thread are being overly dismissive about how useful LLMs are, just because they're associated with techbros who are often associated with relatively useless stuff like crypto.
I mean most people still can't run an LLM on their local machine, which vastly limits what developers can use them for. No video game or open source software can really include them in any core features because most people can't run them. Give it 3 years when every machine has a dedicated neural chip and devs can start using local LLMs that don't require a cloud connection and Azure credits and you'll start seeing actually interesting and inventive uses of them.
There's still problems with attributing sources of information but I honestly feel like if all LLMs that were trained on copyrighted data had to be published open source so that anyone could use them it would get us enough of the way there that their benefits would outweigh their costs.
It's absolutely a meaningful distinction. Search engines push people to tour website where you can capitalize on your audience however you see fit. LLM's take your content, through them through the mixer and sell it back to people. It's the difference between a movie reviewer explaining a movie and a dude in an alley selling a pirated copy of the movie.
A) An LLM does not inherently sell you anything. Some companies charge you to run and use their LLMs (OpenAI), and some companies publish their LLMs open source for anyone to use (Meta, Microsoft). With neural chips starting to pop in PCs and phones, pretty soon anyone will be able to run an open source LLM locally on their machine, completely for free.
B) LLMs still rarely regurgitate the exact same original source. This would be more like someone in the back alley putting on their own performance of the movie and morphing it and adjusting it in real time based on your prompts and comments, which is a lot closer to parody and fair use than blatant piracy.