I have really mixed feelings about this. My stance is that I don’t you should need permission to train on somebody else’s work since that is far too restrictive on what people can do with the music (or anything else) they paid for. This assumes it was obtained fairly: buying the tracks of iTunes or similar and not torrenting them or dumping the library from a streaming service. Of course, this can change if a song it taken down from stores (you can’t buy it) or the price is so high that a normal person buying a small amount of songs could not afford them (say 50 USD a track). Same goes for non-commercial remixing and distribution. This is why I thinking judging these models and services on output is fairer: as long as you don’t reproduce the work you trained on I think that should be fine. Now this needs some exceptions: producing a summary, parody, heavily-changed version/sample (of these, I think this is the only one that is not protected already despite widespread use in music already).
So putting this all together: the AIs mentioned seem to have re-produced partial copies of some of their training data, but it required fairly tortured prompts (I think some even provided lyrics in the prompt to get there) to do so since there are protections in place to prevent 1:1 reproductions; in my experience Suno rejects requests that involve artist names and one of the examples puts spaces between the letters of “Mariah”. But the AIs did do it. I’m not sure what to do with this. There have been lawsuits over samples and melodies so this is at least even handed Human vs AI wise. I’ve seen some pretty egregious copies of melodies too outside remixed and bootlegs to so these protections aren’t useless. I don’t know if maybe more work can be done to essentially Content ID AI output first to try and reduce this in the future? That said, if you wanted to just avoid paying for a song there are much easier ways to do it than getting a commercial AI service to make a poor quality replica. The lawsuit has some merit in that the AI produced replicas it shouldn’t have, but much of this wreaks of the kind of overreach that drives people to torrents in the first place.
My take is that you can train AI on whatever you want for research purposes, but if you brazenly distribute models trained on other people's content, you should be liable for theft, especially if you are profiting off of it.
Just because AI has so much potential doesn't mean we should be reckless and abusive with it. Just because we can build a plagiarism machine capable of reproducing facsimiles of humanity doesn't mean that how we are building that is ethical or legal.
Copyright infringement becomes theft when you make money off of someone else's work, which is the goal of every one of these AI companies. I 100% mean theft.
Interesting article, but in my experience it overstates the problem... at least for Facebook itself (I have zero interaction with Instagram, Threads, or VR).
I've gone back to Facebook for the last few months, and out of what it mentions, I've only seen like half of it, mostly in the comment sections.
Or to be more precise, for 2024 Q2, I'm seeing:
election disinformation - almost none
violent content
child sexual abuse material
hate speech - only in comments
fake news - almost none
crypto scams - a few
phishing - a few
hacking
romance scams - almost none
AI content - almost none
uncanny valley stuff
The article however forgot to include:
science deniers - a lot in open comments, very few in groups
religious zealots - in comments
political trolls - a few in comments
state-sponsored propagandists - a few in comments
general trolls - a few in comments
Still interesting how I get close to zero of these in my main feed.
there’s a level of disinvestment in Facebook
Disagree. Facebook has reached a "plateau of stability" where the current moderation tools keep enough people on the platform to make it profitable.
I've been actively reporting+blocking problematic content, and while about 99% of my reports end up in "no action was taken", it works wonders to keep my feed and group comments clean.
I found this article quite interesting, as I deactivated my main Facebook account around the time the article asserts Facebook was still "trying" and only recently created a new account under a generic pseudonym to access all the community and small business information that is still locked entirely to the platform. Because I have basically nothing in my feed on this account, Facebook backfills it with "recommended" posts and I was pretty shocked at how universally terrible they are. I guess the algorithm uses my location and gender to generate these recommendations, since I've provided very little in the way of alternative information or interaction for it to use. As a result, my default feed is basically just a wall of misogynistic and highly sexualised slop and even the few genuine recommended posts (like backpackers looking for travelling buddies) are clearly being recommended because they feature young women with a bunch of older men thirsting over them in the comments.
Because I have basically nothing in my feed on this account, Facebook backfills it with "recommended" posts and I was pretty shocked at how universally terrible they are. [...] since I've provided very little in the way of alternative information or interaction for it to use
There is your problem.
When an information-hungry platform like Google or Meta asks you to fill out your preferences "to serve you more relevant content"... they are not lying. I mean, it's also to select ads that will pay more for your attention, but the thing with the content algorithm is, if you don't give it data, then it will ass-u-me that you're statistically most likely to engage with content that is getting most engaged... by people who have also not provided it any data.
The problem with that cohort, is it not only includes the few people with legitimate security concerns, but also those who got dark secrets to hide, and/or are using "incognito" browser mode to look for porn.
I don't like to give too much info about myself, but I also don't want to get stuff intended for the "average horny fanatics" group, so I try to give enough data for the algorithm to put me into a group that makes more sense to me.
And it works. The strongest signal you can send to the algorithm, is blocking content you don't want to see. It's amazing how quickly modern algorithms learn to avoid showing me most porn, politics, or religious content, and instead show me science and humor. They still send like 1% of trash my way, clearly checking whether I'll maybe engage with it, but report+block works wonders.
It's not a "problem", as such. As I said, I created the account to view the pages and groups of small businesses and organisations that have no other online presence. I don't use it for the doomscroll algorithm. This was just my observation of what kind of content is targeted towards males in my location by default.
But it's not exactly a "default", it's more of a "demographic with little data"... and I bet it's small enough that the algorithm is showing exactly the content most of its members are looking for. It's somewhat of a sad reflection on the state of privacy, when keeping things private becomes a segmenting parameter.
Really great article, and thanks for posting the text of it.
Facebook is weird for me because it triggers my FOMO, but then if I use it all I see are a ton of random things with the most toxic people in the world living in the comments.
And similarly I just realized why my friends on instagram use stories and not posts, because for the most part stories is the only place I see content from people I know anymore (and again the FOMO).
I really relate to the sentence at the end, “there are people there but they don’t know why and most of what they are seeing is scammy or weird.”
Really great article, and thanks for posting the text of it.
You're welcome. It was their daily free article to email subscribers. I can't afford to pay a subscription fee for full access but I find the combination of their mailing list + podcast is a good way to keep up to date with their investigations.
I left Facebook in 2014, having had to rejoin because in that era, you had to have an account to get a job. Which is another topic but worth keeping in mind.
If I don't know why I'm somewhere, I leave. Rave, website, bar ... these are all the same questions, just with less external pressure because you aren't the product in the other two situations.
Damn, they really overfit their music models. With image generation and text prediction it's very hard to prove a direct connection, but with four or five of those songs it's unmistakable that the original songs were used to generate the music output
I wonder what the effect will be of fixing the models' overfitting. I'm guessing it'll generate worse music, or they would've done so already.
Quite sad that it took the music industry to notice before any lawsuits with a chance of succeeding got off the ground.
That is a uniquely awesome hed. And only strengthens my belief that 404 Media is going to make corporate journalism wish that they'd not shit the bed to the extent that viable alternative options sprang up.
Yeah. Imo their one hope was to make trackers that leverage both networks, as a sort of middle ground device for like a family with 1 android and 1 iPhone. Without that pivot they've seemed dead since the airtag launch.
Targeting an open source project. So brave, what a statement. /s
This has 0 effect on all the big AI companies like OpenAI, Microsoft, Google, etc. All this does is make it harder for FOSS projects and leaves the corporations to dominate.
They are wrong. Theft means depriving someone of having something, and that’s not the case here. It’s more a “they’re taking our jobs” kind of situation.
You forget all the images that "AI" models are trained on without consent or payment. Plus as you say, that training could result in the same artists losing work. Double theft, of IP and future income.
Important difference between you and an ML model: you can enjoy that art (YMMV), the ML never will.
There is a similar distinction between artists and galleries putting artwork to the public, and corporations auto-scraping billions of artwork for a statistical engine to mass produce qualitatively lesser versions.
Paywalled. Here's the article via the Universal Summarizer by Kagi.
Google has experienced thousands of privacy incidents and security issues over a 6-year period from 2013 to 2018, according to an internal database obtained by 404 Media.
The privacy incidents range from small issues like a single errant email containing personal information to substantial data leaks and impending raids on Google offices.
The incidents involve Google's own products, data collection practices, vulnerabilities in third-party vendors, and mistakes made by Google staff, contractors, or others impacting Google systems.
The incidents include Google accidentally collecting children's voice data, leaking the trips and home addresses of carpool users, and making YouTube recommendations based on users' deleted watch history.
While individually the incidents may have only impacted a relatively small number of people, or were fixed quickly, collectively they show how a powerful company like Google manages and often mismanages a large amount of sensitive personal data.
Google employees internally report these privacy and security issues, assigning them priority ratings from P0 (highest) to P1.
The database obtained by 404 Media contains thousands of reports of these incidents over the 6-year period.
The revelations highlight the challenges major tech companies face in protecting user privacy and data, even with internal reporting systems.
The incidents suggest Google may not always be fully transparent about privacy and security issues impacting its users.
The article suggests the need for greater scrutiny and accountability around how large tech companies like Google handle sensitive user data.
That's what happens when people who don't understand LLMs try to profit off of LMMs. You can be sure that the actual programmers at Google told corporate that this would be the exact thing that would happen, but corporate pushed ahead anyway. The programmers shrugged, did their job, and brushed up their CVs just in case.
404media.co
Newest