I did read the article. At least the 3 paragraphs the web server chose to show to my user agent. I believe my statement is consistent with these 3 paragraphs. There is no information about how accessing public media from a public server and doing some computation on it is illegal under copyright as it is currently. Many cases by authors have already been thrown out re: training of non-book LLM on book contents.
Indeed, if we take the US government's established behavior re: computer analysis, it doesn't even count as looking at something (like NSA spy records) unless a human looks at it. Computers can compute however much they want on data. It only matters when humans are shown the otherwise illegally gathered (or copyrighted) data directly.
If there are more than 3 paragraphs for your user agent perhaps you can scrape it and post it here as a reply.
>> A former Nvidia employee, whom 404 Media granted anonymity to speak about internal Nvidia processes, said that employees were asked to scrape videos from Netflix, YouTube, and other sources
Most videos on Youtube are wide open and fair game. The US Supreme Court categorizes the site as "gate up" with no protection of content requiring logins or anything of that nature - if you request the content from the server, the server gives it to you. YouTube has gatekeeping mechanisms in play, allowing sales of content requiring login. Those are bound by ToS.
NetFlix is gate down. That means use of the content is subject to jursidictional relevance of terms of service - in various parts of the world, this means different things, but assuming it's in the US, that means violations of the ToS can result in Netflix shutting down an account - anything beyond that would require a crime of some sort to be committed. Training an AI model would seem to be protected fair use.
Having copies of streams, and other technicalities are supercilious garbage gotcha laws. Let's disregard these for the moment. They serve to allow the exploitation of individuals.
NVidia would be guilty of training an AI system on videos which have been paid for, ostensibly through Netflix's system of revenue distribution based on streams and so forth.
I don't see a moral or ethical problem with this. I see a regulatory and legal problem, wherein technology has far outstripped the relevance of laws that are founded in quill and parchment era principles, barely updated to handle publishing of physical media, and are completely unequipped to handle digital data protections and reasonable individual rights. There's nothing bad faith about what NVidia is doing. There are vested interests in demonizing AI, because it threatens a bunch of concentrated wealth that only exists because it can exploit artists at scale, and sell to consumers at grossly overvalued prices.
We need a complete overhaul of copyright and intellectual property laws, and the MAFIAA needs to be shut out of the conversation. No more gotchas, no more empowering crappy people to exploit consumers and artists to nobodies benefit but their own, no more of this notion that people don't own the data or devices they pay for. Lots of really shady, unethical games are played to maintain the current copyright regime, and it's not rational or good for anyone. Laws should serve the society in the context of the people living there. They shouldn't be tools of exploitation wielded by the wealthy.
Such a weird space, talking about the morals of AI scraping/learnings etc...
There is likely very few people in tech who have never pirated software/game/music.
(BBS, phreaking, warez - all got me started in the 80s 90s etc)
The thing is that the scraping by large capable MegaCorpo results in massive profits for them, though those took massive resources.
My only opinion on anything AI WRT data/scraping etc -- is I need a [DELETE_ME] button.
Personally, I think there should be an AIDIDIA
A.I. Digital Identity Data Independence Act.
The ability to remove my data from any company that may profit from it. - Not the hand-waving BS that is the current "opt out" -- Since AI can scrape me so easily, I should be able to wipe my-vectors just as easily.
Surely everyone just say "oh but is anon/normalized etc." Nope. Its not. Never has been.
I should be able to ask any model/system/service that I am interacting with and paying for exactly what it knows about me. How much its profited from me. Give me my full file. I ask AIs all the time "Give me a detailed dossier on company X include ceo, tenure, salary, spouse, (Asking for spouse is important in high profile folks because power couples often have different last names and you get surprised to see whomever else's ear/interests they carry)
--
If we are arguing that AIs cannot scrape massive public data sets - then we am I not allowed to make my self a private dataset.
I actually wrote a white paper about this back in the early 2000s - but it was large centered around portable gaming avatar/personal data... never pursued anything about it - but this is a serious issue.
EDIT: everyone got flagged while I was writing this... why?
It seems to me, it would REQUIRE Megacorps datasets to store the information you don't want them to store in order for them to confirm to such a request.
What if I should be legally within my rights to be able to ask for my company/AI-held-dossier.
And I should be able to ask for such from ALL companies that have an AI dossier about me - and I should be able to confirm/deny/delete/update/[litigate-where-appropriate] for/against my data. But to do so as seamlessly as logging into whatever my profile is that already has my MFA/token/key/whatever - and given that me being the account holder - should be able to see my whole file.
but yeah its not simple - but necessary.
Remember - we aren't even a decade deep into AI Life-Long-Entanglements... so we better get on something asap... instead of clippy and war and politics.
(The strange thing is - that you no longer have an SSN, as it were, you have an LLM (LifeLog has arrived)
---
@kklisura:
I want GDPRGPT+
Seriously - I I want it on steroids and I want DarkPatterns to be illegal when it comes to such GDPR and the recent law for certain unsub rules in the US are not strong enough. And that third party data brokers masking as removal companies profit massively (look at YC's own Opterly - still shady.)
16 comments
[ 3.4 ms ] story [ 46.2 ms ] threadIndeed, if we take the US government's established behavior re: computer analysis, it doesn't even count as looking at something (like NSA spy records) unless a human looks at it. Computers can compute however much they want on data. It only matters when humans are shown the otherwise illegally gathered (or copyrighted) data directly.
If there are more than 3 paragraphs for your user agent perhaps you can scrape it and post it here as a reply.
Most videos on Youtube are wide open and fair game. The US Supreme Court categorizes the site as "gate up" with no protection of content requiring logins or anything of that nature - if you request the content from the server, the server gives it to you. YouTube has gatekeeping mechanisms in play, allowing sales of content requiring login. Those are bound by ToS.
NetFlix is gate down. That means use of the content is subject to jursidictional relevance of terms of service - in various parts of the world, this means different things, but assuming it's in the US, that means violations of the ToS can result in Netflix shutting down an account - anything beyond that would require a crime of some sort to be committed. Training an AI model would seem to be protected fair use.
Having copies of streams, and other technicalities are supercilious garbage gotcha laws. Let's disregard these for the moment. They serve to allow the exploitation of individuals.
NVidia would be guilty of training an AI system on videos which have been paid for, ostensibly through Netflix's system of revenue distribution based on streams and so forth.
I don't see a moral or ethical problem with this. I see a regulatory and legal problem, wherein technology has far outstripped the relevance of laws that are founded in quill and parchment era principles, barely updated to handle publishing of physical media, and are completely unequipped to handle digital data protections and reasonable individual rights. There's nothing bad faith about what NVidia is doing. There are vested interests in demonizing AI, because it threatens a bunch of concentrated wealth that only exists because it can exploit artists at scale, and sell to consumers at grossly overvalued prices.
We need a complete overhaul of copyright and intellectual property laws, and the MAFIAA needs to be shut out of the conversation. No more gotchas, no more empowering crappy people to exploit consumers and artists to nobodies benefit but their own, no more of this notion that people don't own the data or devices they pay for. Lots of really shady, unethical games are played to maintain the current copyright regime, and it's not rational or good for anyone. Laws should serve the society in the context of the people living there. They shouldn't be tools of exploitation wielded by the wealthy.
There is likely very few people in tech who have never pirated software/game/music.
(BBS, phreaking, warez - all got me started in the 80s 90s etc)
The thing is that the scraping by large capable MegaCorpo results in massive profits for them, though those took massive resources.
My only opinion on anything AI WRT data/scraping etc -- is I need a [DELETE_ME] button.
Personally, I think there should be an AIDIDIA
A.I. Digital Identity Data Independence Act.
The ability to remove my data from any company that may profit from it. - Not the hand-waving BS that is the current "opt out" -- Since AI can scrape me so easily, I should be able to wipe my-vectors just as easily.
Surely everyone just say "oh but is anon/normalized etc." Nope. Its not. Never has been.
I should be able to ask any model/system/service that I am interacting with and paying for exactly what it knows about me. How much its profited from me. Give me my full file. I ask AIs all the time "Give me a detailed dossier on company X include ceo, tenure, salary, spouse, (Asking for spouse is important in high profile folks because power couples often have different last names and you get surprised to see whomever else's ear/interests they carry)
--
If we are arguing that AIs cannot scrape massive public data sets - then we am I not allowed to make my self a private dataset.
I actually wrote a white paper about this back in the early 2000s - but it was large centered around portable gaming avatar/personal data... never pursued anything about it - but this is a serious issue.
EDIT: everyone got flagged while I was writing this... why?
It seems to me, it would REQUIRE Megacorps datasets to store the information you don't want them to store in order for them to confirm to such a request.
And I should be able to ask for such from ALL companies that have an AI dossier about me - and I should be able to confirm/deny/delete/update/[litigate-where-appropriate] for/against my data. But to do so as seamlessly as logging into whatever my profile is that already has my MFA/token/key/whatever - and given that me being the account holder - should be able to see my whole file.
but yeah its not simple - but necessary.
Remember - we aren't even a decade deep into AI Life-Long-Entanglements... so we better get on something asap... instead of clippy and war and politics.
(The strange thing is - that you no longer have an SSN, as it were, you have an LLM (LifeLog has arrived)
---
@kklisura:
I want GDPRGPT+
Seriously - I I want it on steroids and I want DarkPatterns to be illegal when it comes to such GDPR and the recent law for certain unsub rules in the US are not strong enough. And that third party data brokers masking as removal companies profit massively (look at YC's own Opterly - still shady.)
Thanks for some great ideas/input.
>I actually wrote a white paper about this back in the early 2000s
If you have a copy/ summary I'd like to see it. For academic purposes only.