part 1. talk about AI

 i saw this afternoon a post about AI being wrong, where AI valuating that independence of paper american more than 90% probability made by ai. it is really embrassing and big mistake of AI assesment. this issue make "very" beginner in AI world or just a pure user, have less faith in AI because that assesment is a datal mistake.

a professional give an explanation because it was written by very good grammar, well structured and veri proper preparation, that make these paper really structured and become like "AI made" because too perfect for human to have paper like that perfect structure. this explanation give a logic answer that can be know by 2 category. for user that explanation make sense, and in other hand also tell the reason behind AI whic is accaptable. despite it's accaptable, doesn't mean we should satisfy and stop in that explanation. this is a very urgent in AI development where they should "teach" to know more a liittle insight, so their response not just rely on user input (yes i know this is very hard to implement since if AI should "multi diciplin" context, but it's the challenge and if they can solve it, AI will be uppgraded in to the next level).

in this article, i try to a give a little explanation why it's really acceptable if a ai give a wrong answer. note: i'm not a really professional in tech, i have knowledge just because only i did a lot research. so practically my explanation below isn't based on my experience in AI development team, buat as an public who did a lot of research.

so here we are. basically, AI trained by a very huge data (a very very huge, for me it's unimagine because they train AI with literally all historic data, from sport, artist, programming, astronomy, literally all field we can imagine that can scintist explained). oke back to the main stream, they trained by a lot of data. AI basically doesn't know what data it's. so there are many company that spesify on labelled data, they labelled all data, it usually sepatae from "public ai company" we know. that company who have job to data labelling, than sell data labelled for AI company popular we know, like chat  gpt (open AI), grock (owned by elon musk), meta AI (owned by mark zuckerberg), and others. 

so what AI company do to all labelled data they bought for a lot of money? they make algoritma to try connect word to word, word to sentence, on others. it call tokenisasion, to make token as i "data identity" that ai can know. the algoritma they designed can simplify explanation, that algoritma in a simple words is a super complex probability calculator. in example, when user input word study, algoritma search words that connect to study, maybe smart, math, science, university, student, school, teacher, student, and many more related words. so main job of algoritma now is to calculate of probability, so it give answer based on the highest probability calculated. that's why if you put a single word to ask to chat gpt, he confuse and ask more context to give answer and sometimes it try to give several different answer try to catch what context you want. 

yes basically AI is word probability theory calculator that work on work. it's simplied explanation, in the reality it far  more complex. but at least what i write should be a good explanation if you want to know "how AI work and find good answer".

why AI become so smart in majority case? because it train with a huge of data and have complext algoritmic design. asking why AI so smart is like ask why calculator so smart can calculate very fast the equation you need. and then the second question, why AI acceptable if it was wrong? because they are rely on two main things, "data for training" and "algoritma to process the probability calculation". if AI wrong, depends on 2 main thing, first human error, many times it happen when you asking something you totally don't lnow and AI give a wrong path and you dive deeper in to the path. but when it becomes ai wrong because it's own error, it still acceptable maybe because luck of data for training foe context you ask so algoritma didn't enough data to calculate probability properly.