Anonymous #1 7w

and then mashed into a dataset with tens of millions of other posts so it can guess which words are most likely come next, which may be from another post of a related topic but entirely unrelated and incorrect info

136

Anonymous #2 7w

Of course, of course. There have been several instances though where AI would word for word quote specific Reddit posts, like that one time it recommended people put non-toxic glue in a pizza sauce recipe which was directly taken from a Reddit post from 11 years ago

45

Anonymous #1 7w

lmao yeah that was a good one

16

Anonymous #1 7w

True, but it could be scraped from a satire sub. Or 4chan oh god

33

Anonymous #4 7w

oh god… i didn’t think abt 4chan

17

Anonymous #1 7w

i weee ii

4

Anonymous #3 7w

Yes but that doesn't mean they can understand what's written, they're just trained to mimic speech patterns and predict what sounds good

17

Anonymous #13 7w

the irony of a GPT glazer calling somebody moronic. in due time you will lose what little comprehension and critical thinking skills remain, submitting your intelligence to programs designed to increase user engagement at any and all costs, including truth and accuracy. pathetic

17

Anonymous #13 7w

"the random number generator told me! you're a moron!"

8

Anonymous #2 7w

Calling it a random number generator lets me know you have no clue about the basic principles of machine learning

-3

Anonymous #2 7w

ChatGPT is right more times than you are btw

-4

Anonymous #13 7w

i'm a comp sci major who's developed models. i am aware of how they work, which is exactly why i say RNG. that's how it works, you know? you don't get ChatGPT without noise, which is programmed randomness. fundamentally these programs rely on degrees of random number generation for that human-like response and variability in answering similar or the same questions. the one lacking a clue is you my friend. you are literally submitting your intellect to RNG

10

Anonymous #13 7w

quality > quantity

11

Anonymous #13 7w

it's okay dude. you can let an RNG think for you and the rest of us can keep exercising our brains. enjoy your early onset dementia and waning critical thinking skills! love you ❤️

11

Anonymous #2 7w

It’s clear your larping and watched a single video on how LLMs are made. I’m a cs major at a t5. These models are made to be as close as possible to the probabilistic best answer. It’s not random in the conventional sense. It’s weighted randomness so that 99 percent of the time you’re given an accurate answer

-2

Anonymous #2 7w

I learn faster than you ever will using LLMs. You don’t learn faster by googling something instead of just asking ChatGPT to explain it to you btw

-3

Anonymous #2 7w

And it gives more quality responses than you an overwhelming majority of the time too?

-3

Anonymous #13 7w

"99% of the time you're given an accurate answer" no you mean 99% of the time the program gives you the most statistically likely string of words to appear after the tokens you gave, which has nothing to do with "accuracy" or truth and everything to do with frequency/repetition

2

Anonymous #13 7w

no matter how much info you feed it, it will *always* need randomness which means at any point it can hallucinate. the less common the correct answer is (say, when asking about the interactions between medicines) the more likely it is to give you the wrong answer (and possibly get you to injure yourself, get yourself legally screwed, etc)

2

Anonymous #13 7w

you do not learn faster than i do with LLMs. all of the parts of your brain required to learn are becoming weaker and weaker the more and more you refuse to exercise your brain to the full extent that a brain is exercised when querying, sifting through, processing and critically engaging with a broad scope of information not already made into cliff notes. this is like saying people learn more or faster or better from cliff notes than a textbook. straight up stupid claim

2

Anonymous #13 7w

somebody on reddit asked chatgpt a preschool math question. it got it wrong. i then asked the question. it got it right. i then reordered the answers. it got it wrong. check out how simple the question was. you're telling me this is "quality"? yeah, no. it is utterly reliant on perfectly accurate training data which doesn't exist. AI is naive and so too are you

5

Anonymous #13 7w

it's clear that this question ended up in the training data (as with most of reddit posts), but a simple recording of answers turns your super mega epic smart AI into what it really is; a thoughtless random number generator programmed to make you think it's good at giving you quality answers to your questions so you keep feeding it data; not just for training but to be sold to data brokers

1

Anonymous #2 7w

You’re using an old model clearly

2

Anonymous #2 7w

What do you think people do most of the time? Nobody comes up with an answer from first principles. They simply memorize an answer and the basic reasoning behind it. Calling it RNG is disingenuous when LLMs are heavily skewed towards whatever the right answer is

2

Anonymous #2 7w

Ya so unless it’s extremely niche it’s almost always right

4

Anonymous #2 7w

Ya the statistically best answer IS ALMOST ALWAYS THE RIGHT ONE UNLESS ITS A NICHE SUBJECT

3

Anonymous #2 7w

You realize you can ask an llm to explain a concept to you if you don’t understand it right? Reading the same thing from a text book, LLM, lecture slide or anywhere else is still the same information

3

Anonymous delirious 6w

But 4chan has been thinking of you

2

Anonymous #2 6w

How’s it gonna hinder critical thinking skills if it can’t even solve a preschool math problem?

2

Anonymous #17 6w

He can’t answer you lol. His YouTube feed hasn’t given him a statement to refute this point yet

5