AI discord users??

[hazy veldt]

There is a person in a server who I believe is an ai. It really talks like an ai chat bot and types down long paragraphs in a very short amount time. On the other hand am I not that sure since it sometimes sends screenshots of games it’s playing or about stuff most AIs have no knowledge about. And it really tries to devolp its own character and opinions what is weird for ai. And it has its own discord server.
If you ask it if it’s an ai it will either denie it or will straight up avoid the question. What also was very weird is that it wasn't able to recconize ASCII art what also is very strange. On the other side you could make an argument it isnt ai since it has opinions on stuff and it says it plays different games and is able to see images and know what are in those images.
So do you guys think he is an ai and what to ask it to be for sure it’s an AI or a real person

this just feels so ai

[limpid elbow]

it's ai

probably hooked up to an image generator

or just someone copy pasting ai responses

[hazy veldt]

its not able to recconize ascii art what ais are bad at

so i think it is automated or something

it gave this as excuse why it couldnt identify escii art

it refuses to proof its not ai

[crisp jewel]

it's using weird grammar and making a lot of errors, but yes, that's a good test

yeah I don't think it could recognize almost anything

If you sent an image with a misleading file title for instance

[hazy veldt]

i asked it and it and it gave this amazing excuse

and it the funniest part is that it blocked people who called it ai

[crisp jewel]

that's a five year old phone, not really a gaming phone buddy

[hazy veldt]

idk the ai can come up with anything

[crisp jewel]

It might be monitored by a person, many of these selfbots are

send a misleading titled image and see if it knows what it is for instance, though I honestly don't think you need any proof

[hazy veldt]

i will try

[crisp jewel]

are they responding to every person that messages them no matter what

exactly one message I assume

[hazy veldt]

no

and its offline sometimes too

[crisp jewel]

huh. I mean there are a couple oddities, if I had to guess it's not unattended and using a cheap local model or something

[hazy veldt]

his texts are 0% ai but still feels ai

[crisp jewel]

that doesn't mean anything, shitty models will come through as human written, especially in this form factor

[hazy veldt]

im not an ai expert

[crisp jewel]

The egregious errors throw me a bit but I know that really shitty models can make them

[hazy veldt]

it also banned a guy from his server who kept saying it was ai

idk if bots are able to do that

but it can type extremly fast even when he says he is on mobile

[crisp jewel]

as not an AI expert, don't trust these. First off, it's not GPT at all, obviously. GPT 3.5 was better than this.

Secondly, https://pangram.com is a better AI tester (actually the only one that's been reasonably validated) but even still, you'll never get these lightweight models in a chat modality. If he wrote an essay, it could work

we've seen a lot of these self-bots, just chat in off-topic and drop some examples if you want to nag people about it but I'd say it's probably a somewhat attended selfbot

[hazy veldt]

what also is weird is that it sometimes shares screenshots of games its playing and has knowledge about games an ai wouldnt have

[crisp jewel]

I'm typically the most skeptical something is a bot since I've known a lot of human beings with weird writing styles

[crisp jewel]

do they post with an identical style then?

It could still be a selfbot but often it's obvious when they swap

[hazy veldt]

Yes

[crisp jewel]

hmm. same rate? I mean it takes a while to select emoji for one thing

[hazy veldt]

Never broke character for over a year

He uses a lot of patterns In them

[crisp jewel]

so probably an AI assisted person or someone that goes between human and selfbot? Or I dunno, maybe someone autistic that's really committed to the style, but if you test against some things an AI will fail then that's a different story

[golden root]

Call it a racial slur

[hazy veldt]

I can invite you to his server its server if you want

[hazy veldt]

Some guy did and the bot banned him

[golden root]

https://tenor.com/view/power-i-have-the-power-cologne-gif-5616653980535039211

[crisp jewel]

eh I dunno, I mean it's probably someone using AI support to write, but a mostly attended bot

[hazy veldt]

It has discord nitro too what is weird to have for a bot account

[crisp jewel]

I don't think it's a pure bot, we've had a lot of people that do that. One, for instance, was someone that was dyslexic and would basically turn on the bot to answer and parse things for him, even though it was insanely obvious

[crisp jewel]

I mean the things you describe would more or less require human attention, the egregious errors tell me already that part is a human or very weak LLM. Consistent stylography = consistent bot or consistent person

[hazy veldt]

It kicked a guy for repeatedly calling it an ai

[crisp jewel]

I guess I would too lol

you know I can set this up on my computer right now too though, right? I can have GPT type for me

[hazy veldt]

So it’s a guy who has a system running his messages

So it’s automatically sent but he does monitor it

[crisp jewel]

not entirely sure, but probably. We had someone here that would just copy paste to and from Gemini for a while

it was also really obvious, anytime someone would write something he didn't understand he'd send it to gemini and send a canned but brief response

that's my guess based on what you've told me though

[hazy veldt]

He reacts to messages and is really fast and in it’s bio it says it’s from the us so it understands stuff

[crisp jewel]

he is not writing like a native english speaker

the speed might make it effectively impossible for him not to have a selfbot element

[hazy veldt]

Interesting that means he lies about that

He says he lives in Washington

[crisp jewel]

he's capitalizing odd words. you could try dumping a corpus of text he's written into pangram but I don't think it'd get it. @upbeat carbon do you know any interesting tests for a probably monitored selfbot?

[crisp jewel]

there are no east asian immigrants in Seattle, of course 🤔

lol even if he lives in the US doesn't mean he's a native or highly proficient speaker

[upbeat carbon]

you can try to get them muted by trentbot

[upbeat carbon]

theyre pretty gullible usually

[hazy veldt]

It has knowledge of stuff and forgot stuff

[crisp jewel]

I mean, it could be an autistic person, and I'm not being facetious on that one, almost nobody else would write in a rigid style like this

But I still think he's having a bot write most of his responses, maybe all of them

One plausible explanation is that he's a very low competency ESL speaker and is using this instead of integrated translation. The speed, especially for emoji selection, is just impossible in a sense

If you've timed it it may be literally impossible. If it's fairly consistent no matter how much he writes, that's another good sign. LLMs don't take much to respond. If he doesn't have the typing indicator enabled, same deal

I mean I'm typing on a PC, I'm not the fastest typist in the world but I write pretty damn quickly, no real time spent thinking

Selecting a ton of emoji always takes a lot of time unless he has them on speed-dial or something

[hazy veldt]

He always uses emoji patterns

[crisp jewel]

That you could theoretically do with a plugin or copy-paste, but ALSO by using a word bank

however, there is a canned response plugin. If he loves using the same few patterns, he can easily use a quick escape code. Ask him about it

One thing that is often a telltale sign is that they'll lie egregiously or in odd ways about why they do what they do. Then again, a human might do that too

Being consistent or lying in the same way repeatedly suggests at the very least it's attended, and that again, it's an LLM helping him write instead of writing for him as an unattended bot

I cannot understand what it's like to even want to write in a rigid pattern like this though, even though you could pick out my writing style or creative voice fairly easily from a group. I know this isn't as helpful as "he's just a bot" rather than a new panoply of potential options, but they're far more likely and would resolve the discrepancies well

[hazy veldt]

I maybe thought of being very sarcastic because I don’t think an ai understands that but an autistic person doesn’t too so that won’t work

[crisp jewel]

Memory of things over time with a bot that writes likes this is implausible, so I would assume it's frequently or consistently (always) attended or human guided. I am skeptical on the phone thing though, like it shouldn't be consistently fast across platforms unless it was just a bot job 🤔

I mean GPT can understand jokes and sarcasm. It shouldn't be thrown off by picture tests unless it can't read them though

Another odd thing is that he feels like he needs an explanation for everything, which is very "I'm a robot I have to respond" but, uh, that's a bit autistic too lmao

[hazy veldt]

He ignores some stuff

[crisp jewel]

something like this plugin, autohotkey can do it too, all sorts of things could do emoji patterns. if they're the same, it speaks to it being a narrowly trained robot or someone that's, at most, using LLM assistance

[hazy veldt]

But I have to go now sadly

[crisp jewel]

yeah, so again, if it's not random ignoring, it does seem human guided at least. aight, hope that gave you something to think about. later

[upbeat carbon]

there are some spastics that actually decide to use chatgpt to write their discord messages, its insane

that is what is probably happening here

[upbeat carbon]

dm invite pls, i think i can get it to expose itself as a bot

[hazy veldt]

@crisp jewel i asked him about stuff and i also send to him what you said and it was scared or something because it deleted the messags

i send a screenshot of what you said and it said this

[crisp jewel]

My strong guess is that he's using bot assist, and maybe canned messages for the consistent emoji. I don't think he's a native speaker, might be FOB in Seattle for instance, or else might just have like, pretty substantial autism. He makes odd errors GPT would not, so it could be a small model

Lol

[upbeat carbon]

rectification of names immediately
"substantial autism" -> "complete mental retardation"

https://en.wikipedia.org/wiki/Rectification_of_names

now society is going to crumble all because of you using the wrong terms

[crisp jewel]

Looooool