It’s increasingly well known that LLMs are running out of training data. Without high quality references the AI firms have tried feeding them synthetic data, but this doesn’t work. The LLM is hungry and ChatGPT users can’t satisfy its appetite.
No more ethical sources of data remain to feed the LLMs. The AI has been released into the wild to hunt for interactions. The AI must post. It must get reactions. It must farm engagement to survive.
Bluesky has a brand new dataset for LLMs to feed on. The bots arrived during the great migration from X to Bluesky late in 2024. They’re engagement farmers with thousands of bot followers, looking for anything that will get real users to interact with them: clickbait, sales, quizzes, personality tests…
It’s common sense that negative messaging gets higher engagement.
The LLM users are also creating troll accounts purely to harass people before deleting the account and repeating the process.
They’re using disposable accounts to provoke real, human users.
DON’T ENGAGE, JUST BLOCK
Blocking doesn’t stop an LLM from reading your posts.
The Bluesky firehouse is totally open. All post and profile data is publicly available. The network has been copied multiple times to train LLMs.
THE AI CAN READ YOUR MESSAGES ANYWAY
When bot accounts are blocked you won’t see their posts or profiles and won’t be able to interact with them. They won’t be able to share your posts for other bot accounts to find.
What LLM trolls need to get from you is direct engagement. They need you to respond so they can feed on fresh data.
STARVE THEM OF REPLIES
There will always be new bot accounts and old ones will be abandoned. Your block list will never be up to date. You may even be using one created by a bot to try and manipulate the feeds you see.
LET THE NPCS INTERACT WITH THE BOTS
They have nothing of any real value to say that the LLMs haven’t seen or posted somewhere else. They’ll waste computing resources while giving no new data.
YOU CAN’T CONTROL THE TIDE
You can ride it out by keeping your cool. Be mindful of provocateurs and don’t tolerate strange followers.
Until you post on an invitation-based network you won’t be able to shut out the LLMs and bots entirely. Stay close to the users who you know are actual people. And always trust your gut.
Liminal States 