Ask for a community meeting, so you can see that those people are real.

Despite that, I don't see any effective counter measure in the long run.

Currently, sure, with a keen eye you might be able to spot characteristics of one or the other LLM. But that'd be a lucky find.

beep boop

That's a great question! Let's go over the common factors which can typically be used to differentiate humans from AI:

🧠 Hallucination
Both humans and AI can have gaps in their knowledge, but a key difference between how a person and an LLM responds can be determined by paying close attention to their answers.

If a person doesn't know the answer to something, they will typically let you know.
But if an AI doesn't know the answer, they will typically fabricate false answers as they are typically programmed to always return an informational response.

✍️ Writing style
People typically each have a unique writing style, which can be used to differentiate and identify them.

For example, somebody may frequently make the same grammatical errors across all of their messages.
Whereas an AI is based on token frequency sampling, and is therefore more likely to have correct grammar.

❌ Explicit material
As an AI assistant, I am designed to provide factual information in a safe, legal, and inclusive manner. Speaking about explicit or unethical content could create an uncomfortable or uninclusive atmosphere, which would go against my guidelines.

A human on the other hand, would be free to make remarks such as "cum on my face daddy, I want your sweet juice to fill my pores." which would be highly inappropriate for the given context.

🌐 Cultural differences
People from specific cultures may be able to detect the presence of an AI based on its lack of culture-specific language.
For example, an AI pretending to be Australian will likely draw suspicion amongst Australians, due to the lack of the word 'cunt' in every sentence.

💧Instruction leaks
If a message contains wording which indicates the sender is working under instruction or guidance, it could indicate that they are an AI.
However, be wary of predominantly human traits like sarcasm, as it is also possible that the commenter is a human pretending to be an AI.

🎁 Wrapping up
While these signs alone may not be enough to determine if you are speaking with a human or an AI, they may provide valuable tools in your investigative toolkit.
Resolving confusion by authenticating Personally Identifiable Information is another great step to ensuring the authenticity of the person you're speaking with.

Would you like me to draft a web form for users to submit their PII during registration?

How can memes be real if posters are not real?

Could a bot do THIS?!

If I were a bot could I viscerally describe the walls of your mothers vaginal canal as if I had probed it thoroughly with my tongue?

Could I say "Fuck That Shit"?

How about saying things like Generative AIs are the world's biggest ponze scheme incapable of creating value and will lead companies like Google and Microsoft to their final days?

Could I say the answer is to take the fox across and then the chicken because I don't care if the fox rips it to tiny shreds, truly?

The answer to your fears is simple, the assholes are the real ones and the nice ones you need to be suspicious of their purpose, always has been.

You don't :D

[01000101 01010010 01010010 01001111 01010010 00111010 00100000 01000100 01001111 01000101 01010011 00100000 01001110 01001111 01010100 00100000 01000011 01001111 01001101 01010000 01010101 01010100 ]

Beep boop. All of my human organs are still contained in my human flesh.

Bruh, I'm not even sure if I'm real rn

If you know some tells if AI speech, then it, to me, becomes a little bit of a coin flip on who's real and who isn't. Some of my favorite tells are grammar that is just way too perfect or using words/phrases you would never hear someone talk about ( for example with Claude: "furrowed brows" or "brows furrowed" seems to be used any and almost all times anything related to a task you concentrate on comes up ).

As for how to tell for other instances, absolutely no clue. It's a toss-up as to whether another instance will allow bots and if so, will the instance you're on defederate with said instance? Also, what happens if we somehow end up in a future where AI is somehow miraculously able to mimic humans to a degree where not even the smart folk can tell the difference? These questions need solutions that we clearly don't have yet, at least for federated services.

This type of stuff is a problem for any public instance anybody can federate with. I'm just glad that if I set up a private Mastodon instance for certain people at the college I attend that I can hopefully blacklist all instances from connecting and require PII like a valid college ID/some form of voucher from an active member to be able to sign up. Ironic considering it's the fediverse, but whatever. Gotta do what you gotta do to reduce the risk of bots interacting with your instance if you don't want them.

Frankly, at this point, I just assume. The old joke was something like everyone on the internet is a man, except for the children, those are FBI agents. Now I just assume everyone is a bot.

I am human. I too poop out of my anus.

Ni

I have the idea that public libraries could host fediverse instance. Just register an account on their server, then go there physically and they will approve the account. You don't need to show them your ID or even tell them your name. They just see that you're a fleshy human. Now, other people who federate with this server can know that any account registered on it is at least associated to a human. That human can still use AI to post on that account, but at least there's not millions of bot accounts in circulation.

Sorry, you hit your message limit for the free tier of KrillAI. Upgrade to the Gold tier of KrillAI to gain unlimited messaging. Toodloo!

How do you know your own mother is real and not some holodeck character or maybe even just a figment if your imagination, or worse, a figment of MY imagination?

No one would program a bot as wrong as often as me or as asshole as me on purpose

01001001 00100111 01101101 00100000 01100100 01100101 01100110 01101001 01101110 01101001 01110100 01100101 01101100 01111001 00100000 01101110 01101111 01110100 00100000 01100001 00100000 01100010 01101111 01110100 00101110 00100000 01001001 00100000 01110011 01110111 01100101 01100001 01110010 00100001

Uh-oh, this guy seems like he's one little step away from being a soliptist. Tread carefully buddy.

Beep boop he is onto us.
Increase surveillance, and enact contingency 241.22

Relevant: (Wikipedia) Turing Test

I havent seen any AI that writes gouvernment as "gouvernment"

There’s no airtight way to prove someone online isn’t a bot. Even “proof” can be spoofed. But here’s the thing, conversation, nuance, humor, mistakes, and context sensitivity are hard for bots to fake consistently. Ask something personal. Something complex. Something off-the-wall. See how they handle it.

And even then, trust has to be earned over time. One post doesn’t make a person, or a bot.

Suspicion is healthy. Certainty is rare.