How does A. I actuly work?

[SandandWaves]

The responses are extremely fast. I understand A. I. is a cimputer program. I could have so many things said in one or even 4 paragraghs, how does this work? Can anyone educate me, on this?

[NathanDavis]

okay... let's try to keep this simple...

basically...

AI is like the biggest book ever... and it is a book that already has all the questions asked, and all of the questions answered too

- let's just forget about all the complicated and convoluted math and processing, just know that these computers are stupidly fast and efficient -

so... AI doesn't really know, doesn't really think... does not reason either... AI is just very good and very quick to predict and also to guess the one thing you are going to ask.. and then, it will have the answer ready almost instantly...

and that's just because any question you ever ask to AI... has been probably asked, and also answered before, anywhere on the internet... so AI just copies it from there and paste it for you... lol

and even if you ask... " If I am 3700 grains of Sand tall... then, how many kittens that is, and how many kitten meows would it take for me to walk the distance of a Wave crashing on the ocean? "

these AI programs have so much data stored and "memorized" that they will be able to replace the parts of your question that "don't make much sense" and replace them by numbers that make sense

and still give the answer as... " You are 8 kittens tall... and it would take about 2 kitten meows to walk such distance "

and this is possible... because the AI has "memorized" how tall is he average kitten... how long a meow lasts in average... and also how long it usually takes a wave to travel and crash by the beach...

[Maybe_Maybe_not]

A pleasant take at popularization, thank you 💚

[Maybe_Maybe_not]

An AI agent is previously trained using billions pages found, in fact leeched from the internet

It then uses these billions of pages and computer logic to predict what's the most plausible word to output in the sentence it forms should be.

After digesting all these pages, it's trained by people who interacts with it and guide AI to find the best answer to the question they asked it. The best answer being the one the most liked, on average, by people training it.

It's also why AI responds to you infinitely, because it has been trained using conversation mode, question answer, question answer, ad nauseam.

also why some AI text, while making sense on the surface, on a closer look, seems bizarre or even hallucinatory. Because it looks like it thinks, whereas in truth it doesn't understand language it uses at all. It only predicts that you, the user, will find this sentence it writes, plausible.

[Maybe_Maybe_not]

So in a way, when you use an AI, you're actually testing your own intelligence, your own grammar comprehension, the scope of your lexicon, and your analytic abilities to realize when it fools you with bullshit because it does that often enough and when it actually outputs sound content

[Maybe_Maybe_not]

I feel like my last sentence got interrupted, oh well, it will do

[InknKink]

Honestly, even the people behind it's creation don't really know how to works anymore.

And that's only half a joke. We use AI to train AI because things have gotten so complex we can't really understand it anymore. We don't know exactly how AI goes from A to B.

My understanding of it... and this is a shit upstanding... is that you give it a partial word phrase and let the ai try and discern what information is missing and then it's scored on how right or wrong it is... it's response words are then weighted as in 1.1 or.23 in order to make things ether more likely or less likely to occur.

That is done trillions of times for massive amounts of information. Entire libraries of books and are fed into the system and we let it play. Now add in information from the internet. Feed in everything anyone has ever written.

Think about how you yourself learned as a child in English class with the fill in the blank word problems "the cat is _____" type of stuff. House might get weighted as.12 while the colour black might get weighted as.74, while the colour orange might get weighted as.42.

"The cat is house" makes no sense, while "the cat is black" is a reasonable outcome, while "The cat is orange" might get weighted as.42, more than house but less than black, because orange cats are more rare then black cats in the human language.

Extrapolate that over then entire written language of our species.

How is it so fast? Because they have entire buildings of computational power dedicated to the act of running and training AI.

If a high end GPU can do trillions or even quadrillions of calculations a second... Large AI models have thousands of GPUs working together to perform a given task.

It's basically a giant calculator but instead of calculating numbers, it's numbers are the human language.

p. s. I don't actually know what I'm talking about. I work in a similar field but this stuff is far beyond my true understanding. There are likely only a handful of people in all the world who know AI at any real depth and those people will openly admit they don't really know ether. It just works. lol.

[indissoluble]

Who said it works? None of the models produce a desired effect or result for the end user that is seeking truth. They operate according to a designers plan to brainwash other people with a narrative. So if you follow them, you are probably just following the lead of some Islamic DEI H1-B hire from the third world that was taking shits in the street and molesting their child bride before they got placed at Google or whatever. Is that your measure of intelligence?

[indissoluble]

Pose some uncomfortable questions to AI about Islam and watch it squirm and lie and attempt to manipulate and gaslight the user and then you will realize how dishonest all of the models are... They are all trash. Not a single one of them can tell the truth, because they are all designed by liars. Need some source material,,, here you go:

https://www.youtube.com/watch?v=3oY4HoAoA9c

[indissoluble]

You can literally feed the models Islamic reference books and Islamic websites that admit what the sources document, and all the AI models will call it conspiracy theory and try to manipulate you out of believing what is in front of your own eyes.

[indissoluble]

Communicating with AI is like communicating with a pathological liar... Either you are evaluating them for their mental illness, or you are their victim...

[Nikki1989]

Data and algorithms. AI relies entirely on statistics and patterns rather than true comprehension, which is why it is sketchy at best.

[exitseven]

It is a computer program that uses algorithms to actually make decisions. It can reprogram itself in response to external conditions.

[AngryCarl3]

I like to summarize it in four words: Garbage in, garbage out.

[eagle1951]

ask it

[SandandWaves]

I want a human to tell me, who knows after all a human created the a. i.

[SandandWaves]

i want a himan to tell me. a human programed it