>Vibejeets are fucking screenshotting twitch chat and feeding it into GPT
top kek, i dont think you could get more retarded

>>106110387 (OP)
>screenshotting and OCRing (using memeGPT no less) a data source which is easily accessible as raw text
Am I missing something or are they actually that dumb?

>>106110598
/vt/ said this retard basically ran his last company into bankruptcy so i'm thinking yes they're truly that dumb

>>106110598
They're actually that dumb. They already have the data in text format and then they save it as an image and use OCR for absolutely no reason. This is the state of Python "programmers."

Also their "methodology" for deciding if someone is a bot is asking ChatGPT.

>>106110598
Yeah if you already can capture the twitch chat element, just read the raw text. Besides, Tesseract is outdated af and would make many errors

>>106110635
Twitch chat is just IRC. They don't need Selenium or any of this shit.

I'm 99% sure jeets are responsible for all the worthless data inputs spoiling their training data but they can't ban them from using them because that'd be racist

>>106110387 (OP)
Lel I remember this guy from Steven "Zestiny" Bonnell's streams way back in the day. Yeah seems about right.

>>106110686
Devin Nash doesn't sound like an indian name to me

>>106110598
>>106110619
I assume there must be a method to their madness. Probably more to the story, like they can't see the actual twitch chat data for whatever reason, only the video stream

>>106111752
Sounds like the name of someone that underpaids his developers and only hire jeets.

>>106111939
no, there's no reason to screenshot.

>>106110607
Thread? I haven't paid attention to vtubers in like 8 months and have no idea what's going on.

>>106110387 (OP)
sars we are good coders sars give h1b!

>>106110387 (OP)
This approach is utter insanity.

>>106110387 (OP)
IDK shit but I know this is retarded

>>106110655
Checked. This.

Holy shit, script kiddies could do irc bots with no vibe help twenty years ago. What the fuck happened?

>>106110598
I've seen this level of idiocy at multinationals mate

>>106111939
If they're running selenium they can just use the API directly. In fact they could just query the chat DOM element instead of screenshoting.

Holy fuck that's so retarded.

>>106110387 (OP)
>we need 50 million more data centers to capture images of text and then convert those images back into text
Our planet is cooked.

>>106110619
>This is the state of Python "programmers."

Don't put this retardation on us please.

>>106112476
It all makes sense now.

>>106111639
Wait really? I used to watch destiny back in the day (when he was le based) who was this guy?

>>106110387 (OP)

>>106113509
>You can't connect to twitch chat and scrape text easily for 15,000 streams
Alright, that number isn't straightforward, sure. But it's certainly easier than screenshotting 15,000 streams worth of chat activity to run it through OCR. Dear fucking christ.

>>106110387 (OP)
>nuscrapers are braindead niggers who are filtered by parsing XHR requests.
its over.

>>106110598
>>106110607
>>106113509
What is this about? Scraping 15k streams looks like state sponsored surveillance to me, JFC.

>>106113940
Reporting on widespread viewbotting activity on Twitch
Which is something people already know, but this is an attempt to provide "concrete" data
We're just here to laugh at his method of doing it to be honest

>>106113509
> You can't connect to Twitch chat and scrape text easily for 15,000 streams

So instead this guy connects to Twitch chat and uses Selenium to take screenshots for 15,000 streams. He's actually this retarded.

>>106111939
Tech illiterates don't understand how computers work. It wouldn't occur to them that screenshots are less efficient than text.

>>106112476
and we need to make sure those data centers are built in the desert in order to ensure maximum inefficiency

>>106110387 (OP)
This is literally like teenager learning python tier shit. Why is this industry so fucking cringe.

>>106111939
You have to be able to see the data, otherwise your browser couldn't display it. If you can see it, the data is there. Obviously excluding anything pre-rendered.

>>106113509
>we are wasting trillions of dollars on data centers, billions gallons of water a year, and throwing away millions of high-end GPUs so retards can run OCR on twitch screenshots because they can't figure out how to scrape raw text
>these people have a high paying job and you don't
grim

>>106111988
>>>/vt/102995443

>>106113509
its literally just irc, thats it

>saar, this is a perfect project for AI!
smart way to get AI onto your CV.

>>106110387 (OP)
>python selenium to capture twitch chat
>taking screenshots to interpret it
i use 'irc.bot' and it does everything they want with like, 1/100th of the resource requirements
what a retard

How do we feel about about the "AI eats white collar work" thing?

Given that everyone involved in ML is not learning welding, carpentry, etc - they sure don't act like the singularity is going to take everyone's jerbs.

>>106118840
Well unemployed neets on /g/ are telling a guy making money from telling ai to do this that it was "wrong".
Kinda says it all.
>noooooooo you can't fire me you won't knoooow about the irc bot ton get the twitch chats!!!
>don't care. chatgpt fireup selenium and start taking screenshot
imagine i put this text on a crying basedjak and a yes chad nordic guy respectively

>>106118895
that dude just had his ad company fail. nice try.

>>106110598
>and we did some other work to identify what was a bot and what wasn't
so.... they used the language model for getting the text from a picture and then did the actual language work some other way???

>>106111939
>only the video stream
twitch chat isn't a video.

>>106110387 (OP)
This reads like someone desperately trying to sound smart, without realizing how stupid they actually come across in reality. Like some kind of midwit CEO of a failing startup.

>>106112290
People who talk about or use gpt are obviously not a programmer at all

>>106119090
They do anything for those loans

>>106113509
There is just no fucking way hes that retarded.

>>106110387 (OP)
>>106113509
>You can't scrape text easily for 15,000 streams
I'm 99.999999% sure Devin said the exact opposite before. I swear there was a video or stream from over a year ago where he mentioned his company recorded chat logs for literally every channel that has 10+ viewers or some shit. It's either on a video on his channel or when he was on someone else's channel or when he was playing dnd but I swear he said it before.

>>106119247
Holy shit I just realized when he mentioned he was recording chat logs for every channel maybe he was already referring to this Selenium screenshotting setup. I know in the video I'm thinking of he didn't explicitly mention screenshots but I know for a fact he said they had logs of chat or were recording chat for every channel with more than x viewers.

>>106113458
This was post SC2. Just some business associate/friend or whatever back then. He would appear with that Dan Saltman tinychat guy .

>>106119031
It is if the streamer overlays it, retard

>>106119587
twitch chat is IRC, retard

>>106113509
>has money for endless OCR
>doesn't have money for a few residential proxies to scrape a bunch of streams from
ok

I'm not some genius programmer or anything, but can't you just save them as a string and put them into a .txt file pretty easily with Selenium?

>>106110598
It's the most general method. Anything else would frequently need maintenance because of site changes or browser changes.
As long as the OS and the browser manage to update themselves and you upgrade the hardware every few years, this method will keep working basically forever.
It's why agentic GUI automation that receives video and outputs keybarod/mouse commands will eventually win over anything else until they implement trusted computing with biometric signals that require that a human be actually present and paying attention to the screen, then we'll go back to paying indians to do data entry.
As a NEET with a 10 year old crusty twinkpad, sure, it makes more sense to do an html automation. When you're a big company with thousands of automations across all your offices and a dedicated team being paid $100/hour to develop them and keep them working, you would rather spend hundreds of dollars just to do that one stupid job if it means nor or minimal dev hours.
t. somebody who works at GUI automation and who's job is to nanny Selenium scripts 8 hours a day

>>106114048
>Reporting on widespread viewbotting activity on Twitch
With the added bonus that they could train the most annoying and retarded LLM you've ever seen with the data
>[Thinking...] PogChamp PogChamp PogChamp PogChamp PogChamp PogChamp PogChamp PogChamp

>>106120279
Twitch chat can be connected to with literally any IRC client because it’s an IRC chat. Use that information how you will
>>106120426
You are the blackest gorilla nigger in the jungle, see above for why

>>106120782
I didn't even think of that, that's how you add the chat plugins to OBS.
Does that mean the same vulnerabilities that existed in old IRC chatrooms are still present?

>>106110387 (OP)
Is there seriously not some json api for chat

>>106120087
He would still need multiple connections/proxies for the retarded OCR method. In both cases he still needs to be connected to the chat...

>>106120087
Even a VPN + those sketchy free proxy lists works for twitch viewbots ime, wouldn't it work for that too?

>>106120827
idk how it works now but it used to quite literally just be an irc connection

>>106120975
still is