charonn0

Remember Valentine's day 2004, when San Francisco county started issuing marriage licenses to same-sex couples?

Thousands of couples showed up. Some from the other side of the planet, some from the other side of town. The County Clerk was overwhelmed and there weren't nearly enough wedding officiants to keep up. So they put out a call for volunteers to be deputized by the Clerk as county marriage commissioners. I volunteered and officiated at dozens of ceremonies at city hall.

Still have my official commission hanging on my wall.

These would be 'rebel' states are among the poorest and most heavily dependent on federal subsidies. They need the US more than the US needs them.

Yes. Maybe not Hitler-level evil, but there's nothing commendable or useful in being a troll. No one besides you gets anything out of it. You're doing it on purpose, and for your own gratification. You're basically at the same moral level as a public masturbator.

It’s great thinking up insults that are as way off the wall as possible and just seeing how people react.

Troll.

I think we should have a rule that says if a LLM company invokes fair use on the training inputs then the outputs are public domain.

Yeah, the headline makes it sound like they had cameras in the toilets or something.

SSL/TLS, the "S" in HTTPS, and other network encryption protocols such as SSH, use a technique called a Diffie-Hellman key exchange. This is a mode of cryptography where each side generates two keys: a public half and a private half. Anything encrypted with the public half is only decryptable by the associated private half (and vice versa).

You and Youtube only ever exchange the public halves of your respective key pairs. If someone snoops on the key exchange all they can do is insert spoofed messages, not decrypt real ones.

Moreover, the keypairs are generated on the fly for each new session rather than reused. This means that even a future compromise of youtube won't unlock old sessions. This is a concept called forward secrecy.

Message spoofing is prevented by digital signatures. These also use the Diffie-Hellman principle of pairs of public/private keys, but use separate longer-term key pairs than those used with encryption. The public half of youtube's signing key, as presented by the server when you connect to it, has to be digitally signed by a well-known public authority whose public signing key was shipped with your web browser.

If OpenAI owns a Copyright on the output of their LLMs, then I side with the NYT.

If the output is public domain--that is you or I could use it commercially without OpenAI's permission--then I side with OpenAI.

Sort of like how a spell checker works. The dictionary is Copyrighted, the spell check software is Copyrighted, but using it on your document doesn't grant the spell check vendor any Copyright over it.

I think this strikes a reasonable balance between creators' IP rights, AI companies' interest in expansion, and the public interest in having these tools at our disposal. So, in my scheme, either creators get a royalty, or the LLM company doesn't get to Copyright the outputs. I could even see different AI companies going down different paths and offering different kinds of service based on that distinction.

I just thought "pirate-friendly" was concise.

tl;dr: The users' comments say that a certain ISP is pirate-friendly. Studios want to use the comments against the ISP (not the users).

Sounds like a fatal problem. That's a shame.

Which is exactly why the output of an AI trained on copyrighted inputs should not be copyrightable. It should not become the private property of whichever company owns the language model. That would be bad for a lot more reasons than the potential for laundering open source code.