cross-posted from: https://lemmy.world/post/51671840

📌 Pinterest used ~500 Million ROWS¹ of user pins² to train it's AI!

Data from 2009-Now, however the Social Media corp claims "users have granted Pinterest a license".

What license? Pretty sure AI training fine print wasn't a thing in 2009!

It even includes "limited personal information".

Grounds for a lawsuit? 🤓

¹ Never heard of this unit ever before, but even if we assume 1 ROW = 4-6 Pins that is 2-3 BILLION user Pins!

² Post would sound way worse wouldn't it?

🔗 Source: https://help.pinterest.com/en/article/ai-at-pinterest

Screenshot of the "Generative AI and model transparency" section on this page: https://help.pinterest.com/en/article/ai-at-pinterest Some sentences are underlined in red. Pinterest Canvas Version 6 (“Pinterest Canvas”) is a multimodal foundation model that powers Pinterest experiences by enabling the enhancement of images. Pinterest Canvas has been trained on approximately 500,000,000 rows of publicly available Pins, Pin descriptions, and Pin metadata that Pinterest users have saved or uploaded to Pinterest, to which the users have granted Pinterest a license, as well as the continuous use of synthetic data. Training data rows may include one or more images together with text instructions and a target image. The data in the datasets were collected by Pinterest as early as 2009 and were first used in the development of Pinterest Canvas in or around January 2023. The data collection is ongoing. The data in the datasets may include limited personal information, and material that may be protected by copyright, trademark, and/or patent. The datasets do not include aggregate user information. The datasets were cleansed, processed, and modified for quality purposes and to meet our trust, safety, and other internal standards.

all 7 comments

sorted by: hot top controversial new old
[–] 10 points 1 month ago (4 children)

¹ "ROWS" probably means database rows/records. Therefore I would assume one row = 1📍.

  • source
  • hideshow 4 child comments
  • [–] 4 points 1 month ago

    Yeah, it's not even capitalized in the source article, so OP just invented a new unit to be 4-6x more outraged, lol.

    Unless it refers to the famous Really Outrageously Wide Stack, usually representing stacks of 16386 records each. Omg, it's over 8 TRILLION pins!

    In short, Pinterest sucks and OP enjoys clickbait writing style. Avoiding both is probably the best way forward.

  • source
  • parent
  • [–] [S] 1 point 1 month ago (1 child)

    Well pins are usually organized in rows of 4-6, when displayed on Pinterest and in a database each row could have a different type of data, therefore it really isn't that clear to me.

  • source
  • parent
  • hideshow 1 child comment
  • [–] 5 points 1 month ago (1 child)

    This post has triggered my outrage!

    ...because this is totally normal database stuff. Like, where's the meat? My phone could handle 500 million rows of data. Is that supposed to be a lot?

    I'm not exaggerating about the phone part: A modern smartphone can handle an SQLite DB with billions of records (aka "ROWS", LOL).

  • source
  • hideshow 1 child comment
  • 11 comments in c/fuck_ai@lemmy.world view that post
    [–] 22 points 1 month ago* (last edited 1 month ago) (9 children)

    Isn´t Pinterest the site nobody visits anymore because it's overrun with AI slop? I think it's dead.

  • source
  • hideshow 9 child comments
  • [–] 9 points 1 month ago (5 children)

    Hmm a nice contribution to Model Collapse

  • source
  • parent
  • hideshow 5 child comments
  • [–] 5 points 1 month ago (2 children)

    The only thing worth a damn is genuinely human-created content, but only because you can sell it to train AI which produces only negative value.

    We are in dire need of a market correction.

  • source
  • parent
  • hideshow 2 child comments
  • [–] 2 points 1 month ago (1 child)

    Oh I agree, LLMs poisoning themselves on their own slop may help. However, the bubble should not pop yet - it needs to hurt a lot more, or people will never learn.

  • source
  • parent
  • hideshow 1 child comment
  • [–] 2 points 1 month ago

    People who control content-sites like Pinterest don't need to wait for the bubble to pop. They should see their eyeball count dropping precipitously the more AI content they have, and the fewer eyeballs means the less value the site has. But capitalism has trouble seeing past the next quarter.

  • source
  • parent
  • [–] 3 points 1 month ago (2 children)

    Your question assumes only one such site exists.

    Pinterest is that way, but not exclusively.

  • source
  • parent
  • hideshow 2 child comments
  • [–] 1 point 1 month ago (1 child)

    I don't know... I've accidentally stumbled on Pinterest a couple of times in the last few years and it's been absolute garbage all the way down.

  • source
  • parent
  • hideshow 1 child comment
  • [–] 3 points 1 month ago

    Nice to know that I was unknowingly contributing to AI training every time I saved an image I appreciated back in the 2010s, too. At least I put the wrong birth date on my account, I think.

    I’m no longer surprised by this shit, but I’m never any less angry. Fuck these people.

  • source
  • 8 comments in c/privacy@lemmy.ml view that post
    [–] 25 points 1 month ago (4 children)

    Why would a row be more than one pin? That's not how computering works.

    And yeah, when you give a company license to do whatever they want with your data, "whatever they want" includes AI.

    I mean, fuck Pinterest, but this isn't news.

  • source
  • hideshow 4 child comments
  • [–] 3 points 1 month ago* (1 child)

    And yeah, when you give a company license to do whatever they want with your data, “whatever they want” includes AI.

    I think an important point is that with the decline of Pinterest's popularity, most of their data was probably acquired many years before genAI became mainstream. (Like I think I only used it around 2008-2010.) So most users probably never imagined their data could be used for training like this in the future.

  • source
  • parent
  • hideshow 1 child comment
  • [–] 1 point 1 month ago

    In data analytics you sometimes see some weird datasets spanning across a whole lot of columns with what would be otherwise normalised in different records with relations instead.

    Shit can be weird in data…

  • source
  • parent
  • [–] 23 points 1 month ago

    A row is just an entry in a database, I have no idea why you would assume each one represents 4-6. It's already bad enough, you don't need to misinform to make it appear worse, you'll only lose credibility.

  • source
  • [–] 16 points 1 month ago

    If they were publicly available pins, and users agreed to the license that gave them permission to use their data, then they can do that.

    I think the broader issue is people being acclimated to permissive licenses, and not having privacy as a default

  • source
  • Are there any good alternatives for pintrest?

  • source
  • 3 comments in c/Pinterest@sh.itjust.works view that post

    So basically they’re admitting that they’re even breaking the copyright law? Man, there is just so much wrong with this. I knew Pinterest was bad, but I didn’t think it was that bad. Shame on Pinterest.

  • source
  • [–] 3 points 1 month ago (1 child)

    Hopefully it made the AI learn the difference between "it’s" and "its"

  • source
  • hideshow 1 child comment
  • 2 comments in c/MadeInDex@lemmy.world view that post
    [–] 7 points 1 month ago (1 child)

    I guarantee you their license from inception has given them control and ownership of data you put on the platform in perpetuity

    This is basically what everyone does and no one cares because it’s buried in a 50+ page tos/eula that changes every few months and apparently is binding if you simply click a button

    The tv stuff is what everyone’s mad about right now but do you know how many photo kiosks and websites reserve the right to use the pictures you upload for printing? How many games and cell phone apps profile you? How many smart devices phone home with local WiFi ssds, usage patterns, etc?

    Once they establish ownership they can exercise control pretty much however they want. Model training is just the new “this will be a cash cow somehow” fad and everyone with a massive dataset is looking to capitalize on it

    We have let the advertising and data collection industry get completely out of control because the overwhelming majority of people are mindless consumers that don’t give a fuck

  • source
  • hideshow 1 child comment
  • 2 comments in c/antisocialmedia@piefed.social view that post
    [–] 11 points 1 month ago

    Pinterest's whole model is locking other people's content behind the walls of its own platform. It should've been illegal from the start.

  • source
  • [–] 1 point 1 month ago

    could 1 mean row in the DB sense here which presumably would be 1:1 with Pins?

  • source