Now that ChatGPT is being trained using Reddit posts
Now that ChatGPT is being trained using Reddit posts
it will loose its ability to differentiate between there and their and its and it’s.
87 comments
-
And it will get LOSE and LOOSE mixed up like you did
-
it will UNLEASH its ability to differentiate between there and their and its and it’s.
-
-
I'm waiting for it to start using units of banana for all quantities of things
-
ChatGPT trained used Reddit posts -> ChatGPT goes temporarily “insane”
Coincidence? I don't think so.
-
This is exactly what I was thinking.
And maybe some more people did what i did. Not deleting my accounts but replacing all my posts with content created by a bullshit-generator. Made texts look normal, but everything was completely senseless.
-
Back in june-july, I used a screen tapping tool + boost to go through and change every comment i could edit with generic type fill, then waited something like 2 weeks in hopes that all of their servers would update to the new text, and then used the same app to delete each comment and post, and then the account itself. Its about all I could think to do.
-
I hope so. I generally liked that idea back then, but couldn’t do that to my historical collection of words. My words remain in the cloud, as they always will
-
-
They have always trained on reddit data, like, gpt2 was, i'm unsure about gpt1
-
I downloaded my content before changing the posts to nonsense.
-
-
ChatGPT 9 to be trained on R9K posts. Won't be able to distinguish fake free text from real.
-
-
ChatGPT also chooses that guy's dead wife
-
The Narwhal Bacons at Midnight.
-
-
On the contrary, it'll becomes excessively perfectionist about it. Can't even say "could have" without someone coming in and saying "THANK YOU FOR NOT SAYING OF"
-
It already was, the only difference is that now reddit is getting paid for it.
-
Would have and would of
-
Would've
-
Wood're
-
-
This one is the worst for me
-
-
Its going to be a poop knife wielding guy with 2 broken arms out to get those jackdaws.
-
From now on, when you say something like "I think I can give my hoodie to my girlfriend", it will answer"and my axe""
-
GROND
-
"I also choose this guy's dead wife"
-
Not always.
Sometimes it will say "and my bow". :-P
-
-
It was already trained on Reddit posts. It's just now they're paying for it.
-
Yeah Twitter and Reddit were probably a huge source of its original training back when their API was free. I'm pretty sure that's what made them lock it up and ruin everything. Now only companies who already have a ton of money can buy access.
...unless they just use a web scraper instead
-
-
And between were, we’re and where.
-
Insure and ensure.
-
-
ChatGPT was already trained on Reddit data. Check this video to see how one reddit username caused bugs on it: https://youtu.be/WO2X3oZEJOA?si=maWhUpJRf0ZSF_1T
-
I'm not gonna watch, but I assume little Bobby Tables strikes again.
-
It's about the counting subreddit. It was used on the token generation database, but then removed on the training. This user posted so much on that subreddit that a token with its username was created, but then it had nothing associated with it in the training and the model dosen't know how to act when the token is present.
-
-
Here is an alternative Piped link(s):
https://piped.video/WO2X3oZEJOA?si=maWhUpJRf0ZSF_1T
Piped is a privacy-respecting open-source alternative frontend to YouTube.
I'm open-source; check me out at GitHub.
-
-
It will also reply "Yes." to questions "is it A or B?".
-
Perfectly acceptable answer
-
Not of it's neither A nor B ;)
Would you trust ChatGPT to know?
-
-
-
Don't forget the bullshit that is "would of"
-
Who could of
-
-
"Can't even breath"
-
"What is a giraffe?"
ChatGPT: "geraffes are so dumb."
-
“I have not been trained to answer questions about stupid long horses.”
-
Would you rather fight a long horse or a short giraffe?
-
-
-
Your right.
-
And then and than.
-
Is it a showerthought if it's actually just incorrect
-
And when it learns something new, the response will be "Holy Hell".
-
TIL
-
-
Sure it might have some effect, but a big part of ChatGPT besides "raw" training data is RLHF, reinforcement learning from human feedback. Realistically, the bigger problem is training on AI-generated content that might have correct spelling, but hardly makes sense.
-
Then I did the right thing by replacing my texts with correct spelled nonsense.
-
-
The same for Gemini, Google brought its api
-
My iphone lost that ability year’s ago
-
I don't know what to think about this, because on one hand I don't like not beeing asked permission, but on the other hand I'm glad that my opinions will be somewhat reflected in chatgpt
-
I’m a bit annoyed by someone else profitting off my words, though. I freely gave them to the world I guess, and never objected to search engines using them, but my words were not “monetized”, even if they were later used to sell advertising. But it just doesn’t seem right for Reddit and others to be cashing in
-
-
it always did though, just now they get paid for it.
-
That will be more annoying then ever!
-
No! AI is going to be my personal hell!
-
It will be 'rediculous' when it confuses breaks for brakes.
-
It will also start responding to prompts with, "Just listen to yourself!" and "Imagine actually thinking XYZ..."
And it'll constantly and petulantly demand sources for everything you type into it.
-
Don’t forget they’re.
-
ah man you're fucking dead on
I spent cumulative hours of my time writing replies that just said "their" etc
-
Per the community rules, try to keep the complete shower thought in the title.
If you forget the rule, its their in the sidebar ;)
-
"Google is free"
87 comments