What a wild take.
Open up access for machines, but keep human access closed off?
In the age of sloppification, you’d think the correct move would be to preserve signal-to-noise ratio, by opening up human access — for read and write.
I hate these arguments that are like “We need to be the ones to ruin everything, cuz otherwise the e n e m y might ruin everything!”
no they’re saying to open access for everybody, loosen up copyright law in general. the reasoning is fucking stupid but thats why politicians might actually listen to it
When I was a wee kid, I thought that scene from the Matrix where Morpehus explains that humans destroyed the whole damn planet just to maybe slow down the machines was stupid.
I mean if you block the sun, we’re all going to fucking die, why would you do something that stupid?
Yeah, well, the last few years has shown that actually at least half the people on the planet would be pro-kill-everything, even if that includes themselves.
So really, this take isn’t remotely shocking anymore.
I think Anna’s Archives point is mainly that given other jurisdictions don’t care about copyright when it comes to training their LLMs, it’s a major and critical disadvantage for countries that do care about copyright for training purposes.
Given they are trying to get political change, it’s likely they think it’s harder to change the status quo for regular people than it is to change it for AI companies. They are still trying get the copyright duration down to 20 years.
A more cynical take would be that Anna’s Archive wants to be able to make money from companies by giving them access to their archive. Maybe they already have a monetary agreement with companies overseas, and want to do the same in the USA.
Uh, they already train on copyrighted material. This whole “overseas” bogeyman is a misnomer, the boogeyman is domestic.
Copyright being more than 10y is a travesty.
I think it’ll quickly move beyond the ‘overseas’ bogeyman anyway, especially with LLMs becoming open-source. Once that gets properly rolling, anybody could theoretically train an LLM on whatever material they want regardless of how illegal it is, and people won’t really give a shit, they’ll just use whichever one works best.
Decent goal, bad reason.
Yeah I agree with copyright just being about 20 years or so, but more just for the sake of art and having a robust public domain that can preserve cultural things. If that also benefits AI in some way, sure, whatever.
I think ultimately there’s always going to be someone who doesn’t give a fuck about copyright anyway, be it China or just some random person on the internet who’ll eventually train an open-source LLM on everything they can get their hands on. The current copyright system hampers innovation (not just in AI, but in all sorts of areas) and holds back culture for the sake of making a handful of corporations rich, and is largely irrelevant nowadays anyway because of the internet IMO.
It’s especially silly because training is transformative use. If the robot read every book in the library, great, that’s what a library is for.
Anything that pushes back copyright is fine by me.
