chatgpt info cannot be just from scraping the web and books

don6644

Junior Member
Joined
Jul 13, 2023
Messages
136
Reaction score
254
while experimenting with chatgpt, I am constantly scratching my head wondering "how?"

chatgpt accurately answers some questions that I cannot imagine have been addressed on forums or in books.

I dont have the exact wording of my question that I asked but I asked something like this question:

my friend keeps asking me stupid questions that I am not an expert on. like he asks me "how long do you think it will take dmv to review my dealer application?" I have no idea. I dont work for dmv. I have never worked for dmv. he has asked me medical, legal, construction questions. I am not a doctor or lawyer or architect. I am tired of his questions. I am tired of going in circles with him.

chatgpt correctly said:
you find it frustrating because you value logical thinking (true)
you are tired of his questions because you feel such questions require expertise that you dont have (true)
you feel uncomfortable because you feel responsible for the answers you give (true, I dont want to tell him something wrong)
he is an outloud kind of thinker. when he is asking these questions, he is just thinking outloud (true, he specifically told me this)
he is not depending on you for expertise, he considers you a smart person and just wants your opinion (true, he specifically told me he considers me smart and just wants my best guess)

the common belief is that chatgpt just scrapes data from websites and books and organizes it. ok fine but how does it conclude this stuff about my friend when provided with the information I gave it? in what forum/book/subreddit did someone ask this question and get such an answer?
 
I think with 7 billion people on Earth and everyone texting in Internet a similar situation just appeared already in history at least a million times. When LLM answers it takes average most “expected” answer PLUS a random bit, which is important, because it makes it more humane. In your case it just accidentally nailed it.
 
this is a common misconception about AI - but the way you described it, that is not how AI works, AI predicts the next word/token based on the previous words/tokens, it is not a simple memorization, it is a prediction - and you can verify this easily, just take a photo that was never taken before and ask what is on the photo.
 
while experimenting with chatgpt, I am constantly scratching my head wondering "how?"

chatgpt accurately answers some questions that I cannot imagine have been addressed on forums or in books.

I dont have the exact wording of my question that I asked but I asked something like this question:

my friend keeps asking me stupid questions that I am not an expert on. like he asks me "how long do you think it will take dmv to review my dealer application?" I have no idea. I dont work for dmv. I have never worked for dmv. he has asked me medical, legal, construction questions. I am not a doctor or lawyer or architect. I am tired of his questions. I am tired of going in circles with him.

chatgpt correctly said:
you find it frustrating because you value logical thinking (true)
you are tired of his questions because you feel such questions require expertise that you dont have (true)
you feel uncomfortable because you feel responsible for the answers you give (true, I dont want to tell him something wrong)
he is an outloud kind of thinker. when he is asking these questions, he is just thinking outloud (true, he specifically told me this)
he is not depending on you for expertise, he considers you a smart person and just wants your opinion (true, he specifically told me he considers me smart and just wants my best guess)

the common belief is that chatgpt just scrapes data from websites and books and organizes it. ok fine but how does it conclude this stuff about my friend when provided with the information I gave it? in what forum/book/subreddit did someone ask this question and get such an answer?
Probably other people have asked the same stuff as you and chatgpt is giving you the answers expected for these questions.
 
this is a common misconception about AI - but the way you described it, that is not how AI works, AI predicts the next word/token based on the previous words/tokens, it is not a simple memorization, it is a prediction - and you can verify this easily, just take a photo that was never taken before and ask what is on the photo.
wrong. that's exactly how ai works. while you are right that it's not memorization it's prediction, the next token is predicted based on what? based on sample data. what sample data? forums, subreddits, books.

Probably other people have asked the same stuff as you and chatgpt is giving you the answers expected for these questions.
I dont know, while I haven't done a search on google, I find it hard to imagine other people have asked this question. my question seems pretty unique. actually never mind, I think there has to be other people out there whose friends keep asking them stupid questions.
 
Back
Top