back
8 comments
Come up with a ~1PB array with a copy of Anna's Archive and/or LibGen, and we can talk.
Would this be illegal to sell or especially to export?
ML weights should not be copyrightable in the US: there is no copyright gained from sweat of the brow. Until it recently became extremely profitable to claim otherwise, pretty much no competent copyright attorney would have told you otherwise.

ML practitioners who make this claim are playing with fire too, since I don't see how its conceivable that ML weights could be copyrightable without being a derived work of the underlying training materials. ML opponents who think arguing for model copyrightability will protect jobs are also playing with fire, since they're pushing for a world where only organizations like google will be able to create powerful ML models by using the TOS to gain access to weights and training material-- there will still be AI but it will be controlled by a few corporations instead of being available to everyone.

Of course, someone could choose to litigate against you any time for essentially any reason. But that's life-- hopefully the creator of this set prices such that they're comfortable with whatever risk they believe they're taking. And in any case, someone offering the same data for download wouldn't be much better off.

Copyright maximalists won't win this fight. They can't be allowed to. If they do, the best models by far will be owned exclusively by those willing to ignore the law. A few rogue states (and/or wealthy nonstate actors) will end up with all the power.

My guess is that copyright is pretty much finished as a concept. Of course the collateral damage as it sinks into the tarpit will be spectacular, but how can it play out any other way?

provokes fond memories of CheapBytes CDs
or maybe:

https://archive.org/details/primetimefreewar0000unse

cool!

https://www.cs.cmu.edu/afs/cs/project/ai-repository/ai/areas...

doesn't seem to have code, just ftp links as far as i poked

lol sorry - the disk is from the 90's era.

I remember prime time freeware and walnut creek cdrom - you could order software on CD because speeds were in baud back then.