Recently, a dataset of 1M Bluesky posts unexpectedly sparked backlash on Bluesky, including accusations of theft, calls for legal action, and death threats. It also uncovered strong anti-AI sentiment on the platform. This write-up reflects on what happened and why. eugeneyan.com/writing/anti/ - ThreadSky

eugeneyan.com • 91 days ago

Recently, a dataset of 1M Bluesky posts unexpectedly sparked backlash on Bluesky, including accusations of theft, calls for legal action, and death threats. It also uncovered strong anti-AI sentiment on the platform. This write-up reflects on what happened and why.

https://eugeneyan.com/writing/anti/

Comments

kashprime.bsky.social•91 days ago

All we need is a tool that blocks everyone that blocked or harassed Alpindale. Every tech-friendly account subscribes, and problem solved.

eugeneyan.com•91 days ago

this isn't as sophisticated but might be helpful nonetheless

https://bsky.app/profile/did:plc:ilhs5ksfpze2l5rqgig4ycl6/lists/3lbxx6amyfk2j

mariaq.bsky.social•91 days ago

📌

nicolay.fyi•90 days ago

I think what we saw in Bluesky, is what we see in the EU. We punish and regulate the good actors, which do it in the open, for everyone, with good intentions, ignoring and empowering the bad, which do it in silence or somewhere else.

eugeneyan.com•90 days ago

i think you might be right, though i hope you're wrong 🤞

nicolay.fyi•90 days ago

Hoping for the best, preparing for the worst

seismowaves.bsky.social•90 days ago

📌

juicyjack.bsky.social•91 days ago

as opposed to what? a pro AI bias? buddy we all left twitter for a reason

libpol.org•90 days ago

many left-leaning people who hate elon are very pro-AI

eugeneyan.com•91 days ago

open and civil discussion, which was what I was looking for as an alternative to twitter

altryne.bsky.social•91 days ago

Great write up Eugene! Sad to see the vitriol and hate that these folks got by using the open platform in the way it was intended to be used all along. Openly!

eugeneyan.com•91 days ago

thank you for the kind words and encouragement Alex 💙

eugeneyan.com•91 days ago

More about the nuances of consent for public social media data

https://bsky.app/profile/cfiesler.bsky.social/post/3lbwurkbfcs2w

eugeneyan.com•91 days ago

How posts are not an extension of bodies and thus norms around consent may not be appropriate

https://bsky.app/profile/hdevalence.bsky.social/post/3lbxz3c3ne22w

clearspark.bsky.social•89 days ago

I took the use of the word "consent" to not be referring to the rhetorical reference to bodies but instead to be referring to the term as used in terms of service (Ctrl+F in any of them for "consent") and other things like that. For example, we all now regularly see the cookie consent banners.

eugeneyan.com•91 days ago

The inherently public nature of the Bluesky firehose API

https://bsky.app/profile/tedunderwood.me/post/3lc3qjuz4as2l

eugeneyan.com•91 days ago

One of the potential benefits of AI

https://bsky.app/profile/howard.fm/post/3lbzo5oarxc2s

eugeneyan.com•91 days ago

The inevitability of Bluesky needing to train its own AI models as it scales

https://bsky.app/profile/dorialexander.bsky.social/post/3lc4av4e7qs2o

eugeneyan.com•91 days ago

The relative insignificance of 1 - 2M posts for training LLMs

https://bsky.app/profile/stellaathena.bsky.social/post/3lbzqnqmo4k27

ikarius.com•90 days ago

In the end, this will be one of the reason why the firehose will be shutdown.
We are not training data for AI.

ikarius.com•90 days ago

Loved the reference to Dune though.

eugeneyan.com•90 days ago

thanks! also, if you're right, i guess the bsky firehose will go the way of the twitter firehose

ericjmorey.com•90 days ago

The firehose is the basic communication method for Bluesky. It seems like you're predicting that Bluesky will eventually shut down or fundamentally change?

ikarius.com•90 days ago

I don’t make predictions. But if the firehose is used in ways that are considered harmful by Bluesky users …

quickinho.bsky.social•90 days ago

https://bsky.app/profile/quickinho.bsky.social/post/3lbyotu72os2m

There’s also this to consider

maxganzii.bsky.social•91 days ago

The conclusion of anti-AI bias seems pretty weakly founded.

eugeneyan.com•91 days ago

oh, perhaps it's just me that isn't used to such reactions to the release of a dataset, and the comments against the training of AI. haven't come across such negative reactions elsewhere tbh

ericjmorey.com•90 days ago

Anti-AI sentiment is strong on many platforms. But people self-filter their communiacation networks over time. On Bsky there was insufficient time for AI enthusiasts to filter out the Anti-AI crowd.

maxganzii.bsky.social•91 days ago

Well, to clarify, it seemed a quite subjective conclusion.

eugeneyan.com•91 days ago

agreed, and it's mostly based on anecdotal, qualitative data. nonetheless, i thought i'd share my analysis. i look forward to others correcting my somewhat pessimistic take.

maxganzii.bsky.social•90 days ago

And anecdotal I would say tends to be under-rated.

badlogic.bsky.social•88 days ago

Interesting, didn't know the EU funded that dataset.

jeremy.lewi.us•90 days ago

I don't think the divide between people who don't want their data "scraped" and those who want to use AI to improve Bluesky are irreconcilable. If your goal is to improve search and discovery and moderation; excluding people who don't want to be included seems fine.

jeremy.lewi.us•90 days ago

We've solved this on web with robots.txt and it seems like something similar for Bluesky could work well. A big question will be whether it's opt out or in.

jeremy.lewi.us•90 days ago

I find it interesting that @jay and others at Bluesky read the room and tried to preempt them by promising not to train models. My assumption is they think there are many legitimate and valuable use cases for AI but they judged scaring people away the bigger risk.

eugeneyan.com•90 days ago

> tried to preempt them by promising not to train models

oh, this is new to me. could you share a reference please? it's hard to imagine how bluesky would scale without models for search, recommendation systems, content moderation, tagging, etc, just to name a few machine learning applications

khalspi.bsky.social•90 days ago

I think this post is what is being referenced

eugeneyan.com•90 days ago

thanks! this is valuable, and i'm glad they specified generative AI so other machine learning systems can still be deployed to serve the community

https://bsky.app/profile/bsky.app/post/3lbvgvbviak2c

khalspi.bsky.social•90 days ago

and this is where we get into the weeds when conflating generative AI and other ML applications

jeremy.lewi.us•90 days ago

I think there is a legitimate concern that AI will take people's data (IP) and use it to compete with them. I think most people in tech can understand this since IP protection is critical to our industry.

Comments

Posting Rules

Reply