if you ask a LLM some a typical human existential question where does it pull the answer from exactly? digested training data like books and scraped websites or other people's conversations? or it's just a random mix of everything.my point is that any LLM should be transparent with their data sources for every answer that's been generated even non-tech related.
Unironically mostly reddit
>>110042252>if you ask a LLM some a typical human existential question where does it pull the answer from exactly?Why, do you think you're the first idiot to ask the same questions? Are you for fucking real, moron?
>>110042252From suicidal people
>>110042252definitely only reddit and not the millions of books it's read
>>110042252It’s a statistical projection. There’s a “space” of possible next words in a sentence and it constructs a sentence based on which words are most likely. It’s not literally pulling from a dataset or anything like that.
>>110042252all the ai companies have document databases with everything they scraped off the internet and the llm will just search from it based on your query
>>110042324Why does it have the writing style of your typical redditor then?
>>110042252A little bit of all of them.>>110042260Claude can't even search reddit, can he? He might be the least reddit AI that exists.
>>110042252I do this in the shower. If I'm going to clean off in the shower anyway I find it more interesting to piss all over myself while I'm nude first. It's not even like a sexual piss fetish, I don't get off on it at all. It's more of a "might as well" kind of thing. I have a "i'll try anything 1000 times" kind of attitude.
>>110042252copy-pasta the most clicked link on google/duckduckgo/bing or made shit up
LLM model training does not require it to be trained on your question specifically. during training the billions of variables are adjusted (not totally at random) and a cost function is run to calculate how wrong the guess was with those changes compared to before and through this method (and with the help of other scary math) the model becomes better and better at predicting text. But you can ask the same model the same question and it'll produce different answers because the token generated is based on a random number roll, appropriately weighted. And things like logical reasoning apply even to novel topics, which is part of why AI used to make shit up so much.
>>110042934I'm gonna fuck you in the ass once, so you have to go do it 999 more times
>>110042252RedditEverything is RedditSnarky personality shite bots
>>110042252>if you ask a LLM some a typical human existential questionoh no no no
