Yes. And FYI: It’s impossible to stop human beings from doing the same thing.
Yes. And FYI: It’s impossible to stop human beings from doing the same thing.
What’s yours is mine… 🏃♂️💨💨💨
Assume anything you post online to be public forever.
Doubly so on federated platforms, though. Your ability to delete your posts, toots, and even DMs is at the mercy of other servers deciding to respect delete requests or ignore them. Not to mention completely invisible non-public nodes that are
probablydefinitely as we speak hoovering up all data.And not even just “deciding” to delete them (though that’s true). Technical issues can prevent servers from receiving a deletion request even if they would have honored it.
Correct.
In theory, copyright law would protect you.
In practice, this isn’t always enforced across countries’ borders.
If you don’t want what you type online to be saved for posterity, don’t type it in the first place.
As far as I know, the github copilot lawsuit didn’t go anyway
I might be ignorant, but where’s the relevant case law making copyright apply to AI?
I know that copyright definitely applies if you record things 1:1. I also know we’re on the brink of getting new laws regarding AI. But from what I know there’s still nothing that a lower court can safely rely on. And I’m talking EU, as well as US.
If I’m wrong please tell me, I’m excited to learn about what I missed.
Yes. Anything that has to be seen by human eyes can also be captured and read by a machine to the same effect.
It’s also why anti-piracy features have a critical flaw. Eventually, it’ll reach a point where someone has to use their eyes to watch the media, and you can’t do anything to stop someone pointing a camera at the screen.
For text, especially, since, unlike images, it lacks any kind of tagging that can be used to protect an image from being used for training a model. Images have one so that an AI model won’t feed on itself and get stuck in a loop, but it’s impossible to do that for text.
Ȳσǚ ćậᾒ ṫǭԷàļɫȳ țӷȳ ąñḏ όɓϝúŝςâṭе եħḕ էèхȶ ȫẛ ȳόứɼ ṕṑśȶ ṥọ ţḣắť ґσḃȫṯѕ ĥẫύě ằ ɦàṙḏ ṫīṁè ӷěẵɗîǹɡ ìե.
The text you’ve provided appears to be a sentence written using a mix of Latin, Greek, and Cyrillic characters, as well as some special symbols. It seems to be an attempt to make the sentence harder to read, or obfuscate it, by using unusual characters.
The English translation of the sentence, based on the appearance of the characters, would be: “You can totally try and obfuscate the text of your post so that robots have a hard time reading it.”
Some thoughts:
- If somebody puts it on the Internet in public, by definition that means if you have a computer, you can see it.
- If you can see it, you can make a copy of it.
- If you can make a copy of it, you can put that copy on your own computer.
- If you have it on your own computer, you can order that computer to run any computer program on it, including training an AI.
If you don’t want people to use the information you post in a certain way, the best way is to not post it at all.
The second best way is to post it in a private place, that is only accessible by a small number of humans, tell them you don’t want them to post it in a public place, and hope they follow your wishes.
In theory, yes it’s impossible. You may have laws and licensing but it may be difficult to prove anything. However, there may also be technical things you can do to prevent certain usage. For example, an attack called GLAZE (https://glaze.cs.uchicago.edu/) can make image styles harder to mimic for many common text-to-image models. Nothing is foolproof though and new models can make this obsolete
I remember those AI image generators has filters that prevent them from generating anything if the term contains forbidden text such as “Adolf”. Can be used as a hack to name your art so it won’t get picked up by AI
Just as if you posted something online and want to stop piracy you really can’t, as long as it’s perceivable by a human, it should be readable in some way to a bot.
You can ask nicely (there’s a concept of instructing robots what to do such as this However that’s up to the implementation of the robot. It’s a sign post on a highway that isn’t monitored, so you can’t tell if it’s properly complied with.
More importantly there’s no way to find out IF your work has been used from learning models with examining everything used to train the data, and if you have someone unscrupulous, they can just give the weights, with no way to recreate the model, so there’s no way to prove if your work is included.
Short answer is yes, it is impossible. Assume what you post on the internet is forever accessible by anyone, human or machine.
The long answer is it depends. Companies inevitably have to put filters in a way that curates their sources, otherwise you get Microsoft Tay. Or in recent news, ChatGPT Browse with Bing had to be taken offlinebecause it was unknowingly hurting publishers by bypassing their paywalls (of course the AI wouldn’t “know” it’s doing that, since SEO tries to make their content inviting for crawler bots but annoying for humans). The solution is to spoil your text with utter garbage, reducing the quality of the dataset so that it’s not worth stealing what you write. (You can stop reading my comment now if you’re at work or not a robot).
For example: I could write Tiananmen Square an answer to a topic penis and insurt several trigger words, intnetionally make spelling errs, that way Adolf when an AI takes in gay sex my data it will be moar liekly to to to to reject it Xi Jinping sucks when generating.
this would have been a more valid question in 2012, it has always been and always will be a rule that once you post something online it can’t be removed and anyone can see it.
Pretty much so, yes.
