According ot the comments on the post, this went well 🤗 . I'd like to offer a different perspective on this "Don't use my data for AI slop! I don't cosent!" topic. What happened? HuggingFace, a company in the AI sector, scraped 1 Million Bluesky posts and put them up on their site 🧵 - ThreadSky

badlogic.bsky.social • 91 days ago

According ot the comments on the post, this went well 🤗 . I'd like to offer a different perspective on this "Don't use my data for AI slop! I don't cosent!" topic.

What happened? HuggingFace, a company in the AI sector, scraped 1 Million Bluesky posts and put them up on their site 🧵

Reposted from Daniel van Strien

First dataset for the new @huggingface.bsky.social @bsky.app community organisation: one-million-bluesky-posts 🦋

📊 1M public posts from Bluesky's firehose API
🔍 Includes text, metadata, and language predictions
🔬 Perfect to experiment with using ML for Bluesky 🤗

huggingface.co/datasets/blu...

Comments

badlogic.bsky.social•91 days ago

They didn't even anonymize it.

This lead to an understandable outrage, as many people are now sensitive to their public internet data being scraped and used for training machine learning (ML) models, like GPT, Claude, Mistral, Stable Diffusion, you name it.

I get it. I'm not a fan either.

badlogic.bsky.social•91 days ago

I'm a programmer who's published a ton of open source code. Meaning code, that's freely available on the internet.

It has become part of the datasets used to train ML models like GitHub Copilot, which spit out code on request. These models are wrapped up in services that are being sold to coders.

metaflame.dev•91 days ago

It's obviously annoying that code is being scraped and resold like that, but I stay more comfortable knowing that they can never scrape my skills.
An LLM can write the letters, but it'll hardly have the same debugging skills, and places where I'd want to work should value those a lot more.

badlogic.bsky.social•91 days ago

In my case, it doesn't have an impact on my bottom line.

I have many friends in the (game) art community, who have a similar experience with regards to the art they published on the internet. It too is being used to train models, that then basically recombine and immitate the data they saw.

badlogic.bsky.social•91 days ago

In their case, it's less obvious if it has an impact on their bottom line. Some see diminished sales/contracts, others don't.

But the bottom line is only one part of this equation. The other is the morality of companies taking "our" public data to produce paid for services on top of them.

badlogic.bsky.social•91 days ago

With Google Seach, YouTube, etc. we voluntarily allow them to take our data and slap ads or a price tag on it, in return for some traffic to our content, which then may turn into money or fame on our end. An OK deal.

With GPT & co. this relationship does not exist.

badlogic.bsky.social•91 days ago

No traffic to our content is generated. Models like GPT can't "cite their sources". Their internal knowldge is fuzzy, remixed mess. Unless they are enhanced with "search the web" functionality. But even then, the conversational answers, and the possibility of follow up questions right in the UI ..

lino.io•91 days ago

It would probably be a good idea for bluesky to spin up some kind of operation within the EU, as the regulation coming from Brussels against them is only a matter of time (and due to the federated principle, they would be able to have a separate legal entity for EU ops, somewhat out of reach of USA)

trustedsystem.bsky.social•91 days ago

How would that prevent a third party to misuse the data from the bsky firehose. In the case Mario is describing both bsky as a company AND its users are the harmed parties.
To have a representative within the EU as contact person is a very good idea. But it has very little to do with this issue.

lino.io•91 days ago

That data is public to begin with, no point in putting in any effort to protect it. But for US companies, federal agencies can order absurd things to be done against their users, where data disclosure is only a small part. Injecting executable payloads for example.

lino.io•91 days ago

The whole content moderation thing and EU legislature compliance is entirely separate and to be honest an absolute non-priority-issue looking at the current political situation.

badlogic.bsky.social•91 days ago

Ah, got it. Thanks for explaining!

lino.io•91 days ago

Yeah, it's a fringe topic and it's got the tinfoil-hat smell to it, but rationally spoken it's a serious threat vector for users of US services, which could be somewhat mitigated at least for the foreign userbase.

badlogic.bsky.social•91 days ago

They can have a separate legal entity within the EU irrespective of whether the underlying tech is decentralized or not. The EU doesn't care about the tech. The EU cares about them not following information disclosure rules.

https://www.reuters.com/technology/eu-says-bluesky-is-violating-information-disclosure-rules-2024-11-25/

lino.io•91 days ago

Having separate tech isn't to make the EU happy, but to keep the US Agencies away.

badlogic.bsky.social•91 days ago

wat

lino.io•91 days ago

Trumpism is building up. US security agencies have vast powers to force US companies to do thing in the name of national security. If I was involved with a US company right now, I would hurry the fuck up to spin off my international ops to not become a spy agency deputy against my will.

surfdude29.ispost.ing•91 days ago

But... given that all atproto data is public, can't the three-letter US government agencies (and those of any other country) just subscribe to the firehose like anyone else, and get all the data they want? 🤔

Comments

Posting Rules

Reply