Registration applications (which are pretty much required on all lemmy servers now) have stopped about 95% of them. Some still get through of course, but the admins of every server are doing hard work to prevent it.

The other 5% are quickly discovered and reported, and they can be banned (and their content removed), at the click of a button.

The LLMs are getting creative with how they answer, and people also are applying with half of the responses LLM based, and half real. But its still not hard to spot them when they give all the same predictable answers.

Its interesting because applications are how the old forum-internet used to keep out bots and trolls, and its still the best method to this day. There’s still work to be done, and it’s a never-ending battle, but I think overall we’re winning and doing better than western surveillance platforms like reddit, youtube, or facebook.

I have to disagree with many of the comments in this thread saying there are no / minimal bots here. I started tagging users pretty frequently instead of blocking over a year ago, and it’s pretty eye opening to see those accounts pop up in the wild posting completely opposing content or opinions from week to week. Sometimes even from thread to thread. Especially when it comes to politics or AI, for example. It doesn’t help that ban evading and making alts is dead simple.

Of course, this is just my opinion based on my own experiment and observations. I’ve been convinced many times I was arguing with some fash chud only to then see them completely agreeing with my position in another thread two days later. Normal people don’t flip flop that easily. I think LLM controlled bots are a totally different beast than what many of us are used to, and its much harder to see past the facade in the moment.

If your fedi client has a tagging feature, you might be surprised what you see when you start paying attention over time.

Edit: Just for clarity, this isn’t some extensive project I devote time and effort to, I don’t have solid answers for you regarding the rigor of my method or why people decide to make or use bots on a niche corner of the internet. I’ve personally noticed a lot of odd account activity in my few years here that has gotten more common with the ubiquity of LLMs. I’m not claiming bots outnumber real users, either. It’s completely fine to disagree! Enjoy the fediverse as it is as you see fit.

From the admin side, we see the following

• lots of accounts trying to register with the same LLM generated text, or nonsensical spam in the registration application. Those get blocked
• some accounts that immediately make it obvious that they are a spam bot by posting unrelated advertise-y content as their first post. Those get banned quickly, and usually the home instance prunes the account within a day

The recent “DM me” spammers and the accounts that post on [email protected] are a good example of the second category.

It’s pretty rare for a bot to be active for much longer than that. Usually some eagle eyed user will spot it and report it up the chain.

Bots are very rare on Lemmy. Some do exist, especially ones that are clearly marked as such by account owners (for various purposes like changing twitter links to xcancel links automatically).

Lemmy/piefed are still pretty resilient against bots and llms since most instances now require an application with short questions that often trip up bots into revealing themselves.

It also helps that most of our userbase is still real, so any bots that get through often stand out and get reported fast, and most instance admins collaborate to help spread the word if one is found to deal with it.

However, there may come a time when it becomes harder and harder to suss out bots in applications or in conversations, at which point we’ll likely need to adopt new ways of verifying that new applicants are human.

Bots outpaced humans officially online like 3 months ago. They are everywhere, consuming everything and generation things at an alarming rate.

That all being said, it seems lemmy isn’t nearly as bad as everywhere else so far.

I don’t notice any botting at all. There’s no karma to farm on Lemmy and there’s no account needed to scrape it so that might be part of it.

I wonder if there’s even any incentive to do botting on Lemmy.

Oh it’s slow alright.

There are not many bots here. So far in the few years I’ve been here, I’d found like three or four accounts that I could reasonably confirm were bots. It didn’t take long until they got banned for inauthentic behavior, especially when providing evidence like screenshots and such.

Overall, still leagues better than reddit which is pretty much 70% bots by volume nowadays.

Just wondering from some of the comments, some people either feel like really drone minded having a hard time to accept the sun is blue just because.

I am trying to figure out what this means

I saw a lazy spam bot that posted the same “Look at my snap, USERNAME” with a stolen pic of a scantily clad woman, that posted on a load of communities (none of which made sense to post on, like one on LinuxGaming)

But I’ve never seen any type of bots more sophisticated than just basic-ass spambots on Lemmy and Piefed.

And the basic-ass spambots posts last for at most 1hr before getting nuked by the mods

They should implement bot detection that looks at the volume of posts and duplicate posts posted across the whole fediverse from one account, bayesian and LLM based filters to detect spam programmatically, PoW to slow down bots which get through.

As well as checking for signs of automation (Puppeteer/Selenium/Playwright) and suspiciously fast or perfect mouse movement in the web client, and showing a captcha to post/comment in the mobile client.

I don’t think most instances do any of that at the moment, and what looked to be a fake Gofundme ended up on the front page, from an account which was absolutely spamming Mastodon relentlessly.

They exist. It’s not even a quarter as bad as reddit