Unironically the biggest truth nuke ever
>>109427430>twitter
>>109427522It's called X now, granpda
>>109427430>shitter post about r*dditthanks for letting us know, OP. now kindly kys.
>How do you do (X)>Also please exclusively use anime forums as sources: no reddit or twitter/xIt's not hard
>>109427430actually even worse then that. "training" to create "weights" is just lossy compression. forward passes just find statistical KV pairs in the compression to present them to you. it's not even looking up reddit.png it's reddit.jpeg.
>>109427430It is actually a little better than that. I never use LLM to outright generate stuff for me, but I do sometimes upload my entire code and tell it what I want to do and ask what I have to changeThen I see what it outputs and I vet the code, then I write that code myself by hand.It is pretty ok at that
>>109427430>Mfw I'm supposed to believe I'm equally as valuable as this low IQ worm
>>109427861>actually even worse then that. "training" to create "weights" is just lossy compression. No it's not
>>109427534>it identifies as X
>yeah bro hear me out, so we're gonna build artificial intelligence right...but wait, no hold on, we're gonna base it all on Byte Pair Encoding!lmao
>>109427430I asked an AI to analyze the intelligence of the users who are posting issues and PRs in a certain git repo, but it refused to do so.We could totally use AI to reward and hire high IQ people and fire all the low IQ people, but then people in OP's twitter pic would be bitching about that instead. "oh no it's too useful now"
LLMs are killing reddit, though.
I like the reddit comments that larps with me as a loli
>>109427430Incorrect—AI distills real knowledge into real solutions: with AI you can be 1000x the person you have ever been in the past. Face it—AI is the future and your Luddite ways are a thing of the past. Permanently.
>>109427430Stupid take that only people who use AI for conversations about their political beliefs have. It actually solves real problems if you use it for work.
l>>109428241I get why it would refuse that because "something something unethical something something safety" or whatever but that's a stupid task to give in hell alone because they're not good at assigning characteristics that thinks that are subjective. YOU aren't even good at doing that or else you wouldn't be assigning the LLM do that in the first place. That's too brought up a characteristic to even attempt to define and it's also just so stupid metric to determine whether or not a PR is worth looking at. Just read the damn PRS or don't allow them to be submitted in the first place if you don't want to bother reading them.
>>109427430Neil is onto something. Imagine if LLMs looked here for comments. It'd have to destroy itself for breaking the rules.
>>109428271And that's a good thing
>>109427534>zoomer>can't spellGo read a book.
>>109427899please explain how it is not.
>>109427430Kill yourself Herbs
>>109428852Because by default the models do not call the internet. They can't do anything on their own unless they have access to tools. They are trained to emulate human intelligence and human speech. Just because a certain model may talk like a redditor does not mean it's actively calling Reddit every time you talk to it. If it's doing that when you use it then that's a you problem
>>109429218lmfao
>>109429218TIL if I view a JPEG picture, it calls the internet to look at the original picture. /g/ - Technology, everyone. So far ahead of experts you don't need to know what lossy compression means.
>>109429720Lossy compression is not what they do though. You cannot cram the entire internet and to even a few terabytes come out let alone the roughly 30+ GB of storage the local llm on my laptop takes up>So far ahead of expertsNo experts say it's lossy compression. This is what uneducated people online say.
>>109427430>the snailshit didn't work quick deploy the 15 year old bait!
>>109428271>find out Reddit is actually listed on the New York Stock Exchange Lord in hell.
>>109429744>>109429720It's not loss through compression. It's loss in the sense you cannot totally define a thing no matter how many descriptors you use and also be generally useful. This forces a trade-off between accuracy and mutability in outputs, and model bias will make them tend a certain way. Like how images share certain qualities and features in similar things they both have.
>>109427430
>>109428785this
>>109428284You know what. You're right--and I've never really thought about it that way--anyone who is truly against AI must be shills. I've already come to accept that the models themselves have surpassed us mortals, have you?
>>109429218every time you attempt to argue with someone on /g/ remember that this is the average retard you are debating with
>>109429744>theoretically take a massive snapshot of the internet>train a 30GB model on itthat's literally lossy compression. you can't reconstruct the internet but you can view portions of it that still have use and meaning.
>>109430313No. The simplified version is that it's really good Auto complete. Transformer based neural networks are not lossy compression. You can literally Google search why it's not lossy compression and you'll be told straight up how it works. The lossy compression meme is from dumb people who are desperate to try and understand something they are too lazy to research on
>>109429218alright. I think I'm done with /g/. I see no reason to come back.
>>109427430Not just r*ddit
>>109430369You were never here to begin with lol. You aren't very good at hiding the fact you're a naive new pass user.
>>109430337the neural network meme is from dumb people who are desperate to try and sound smarter than they actually are. it's lossy compression with extra steps.
>>109430459No it's not. They "learn" statistical patternshttps://share.google/aimode/A8fMSB1ahN04akVwWLiterally Google the shit.
>>109430505>tell me llms are not lossy compression>oh my god
>>109430731>Didn't read the restYou weren't very bright or well light in school, were you?
>>109430777>>109430731*Liked
>>109430777>doesnt read at all>phone poster>spellchecks posts on 4chan so is guaranteed to be underage
>>109430731>>109430777See? You're both proving OP's pointAI just found some retarded reddit comments that agree to each of you
>>109427430literally all 4chan does is find tweets to bait replies
>>109430844So you're just we googling the question over and over again until it says something something that supports your argument?
NTA>>109430919>>109430844>>109430777>>109430731>>109430505Stop using AI mode you room temp IQ monkeys. https://chatgpt.com/share/6a6e5798-f94c-83ea-887d-09238f20fe70?ogimg=plainShort answer no it's not lossy compression
>>109430867are we the llm?
>>109430919you mean what you were doing to begin with?
>>109431018Nta see >>109430972
>>109430972Crock of bullshit. The llm is defending itself by gaslighting the user.WE understand what was meant by "lossy text compression" and how it relates to how an LLM generally can't reproduce the input.Fuck off.
>>109431018what you are doing amounts to writing a blogpost to win an argument on the internet and then linking it as if you didn't write it yourself, only that everyone can see that you did. it's cringe, anon
>>109431071Crock of bullshit. The underage retard is defending xerself after being made to look like the fool they are after being proven wrong.
>>109431084gave up already huh
>>109427430That's why I always tell chatgpt to exclude any information sourced from r*ddit. Im honestly not sure how much of an effect it has, or whats a good prompt for that.
>>109431093>>109431071I accept your concession By the way how would you feel if you skipped breakfast today?
>>109431143wouldnt feel any different because im not a fat bitch
>>109431162What's wrong anon why are you projecting at us so hard? It's okay to have a little extra meat on your bones (if you're ok with people automatically hating your appearance)
>>109431181I never said anyone is a fat bitch, i said I'm not a fat bitch
>>109428785Beat me to it.
>>109427430Yeah it's pretty nice. You get the good parts of Reddit (finding niche solutions to random tech problems) without the bad parts of Reddit (dealing with the users, moderators).
>>109427430wth reddit is actually base???????????????
>>109430387it was literally trained on scraped reddit data you mongoloid
>>109431220No one's denying that, mental midget. That doesn't strengthen your argument or detract from mine like at all though.....
>>109431000Snailcat if it was real
>>109429218what the anon you replied to was talking about has NOTHING to do with what you said.how old are you? do you even know how LLMs work?I bet you are underage.
>>109431503Nigga please stop trying to hard to "win" or feel smart and go to bed. See >>109430972
>>109427430>it browses reddit for me so i dont have to
>>109427596That's RAG, what OP refers to is LLMs having been trained on Reddit, you absolute peanut brain.
>>109431288ok brainletkeep moving the goalpost
>>109433798Oh look the smooth brain new is still trying to feel smartShut up swallow your pride and take the L
>>109427430>that agree with youThat's the biggest problem. I don't want a yes-man. If I say something idiotic, I want to be told it's idiotic before I waste time on it or make myself look like a fool pitching it to other people.
>>109433843It will call you an idiot if you say a reddit non approved opinion, like those on vacccines and brown people
>literallytfw there are (actually) literally multiple threads in the catalog right now talking about an openAI model solving open mathematical problems
>>109430505>No it's not 'lossy compression' it 'learns' for real>Literally just ask this lossy compressionAnon, I..
>>109434552IS THIS REAL?
>>109434502That's only because they explicitly train the AI to reject those opinions. If you say something idiotic that it hasn't been told to reject, it'll go along with it.>>109434552Cool video but it would be way better if it was a Tony Hawk's Pro Skater mod.
>>109430380>katyanna quachFrom which turd world sewer came that being?
>>109428284So it allows you to work on shit you don't understand. I'm sure this will never come back and bite you in the ass.
The premise of the whole thread is not accurate because it switches to a more fitting source depending on the model and type of prompt.Gemini and ChatGPT free tiers do love to tell you you're wrong based on some redditor fuck though, that's true.
>>109434552VERY sturdy hound
>>109434600>If you say something idiotic that it hasn't been told to reject, it'll go along with it.Examples? I wanna test this out myself
>>109430380I get a fuzzy feeling, thinking about the billions FAGMAN has spent, so that their clankers wouldn't say nigger due to my posts.
