eli5: Why does ChatpGPT give responses word-by-word, instead of the whole answer straight away?

537 viewsOtherTechnology

This goes for almost all AI language models that I’ve used.

I ask it a question, and instead of giving me a paragraph instantly, it generates a response word by word, sometimes sticking on a word for a second or two. Why can’t it just paste the entire answer straight away?

In: Technology

28 Answers

Anonymous 0 Comments

Just think of it like this … it’s just a statistics machine. it does not “know” the answer … it just knows what words are more probable to come after what words.

That’s why it’s sometimes sooo wrong.

After it generated a word, it just gets the “best fitting” word that goes the best after that word. After it has 2 words, he works out “what word” fits the best after the next 2 words and so on.

Anonymous 0 Comments

Because all it’s doing is generating what the next work is based on probability. It doesn’t have an entire answer.
Ok, it could generate all the words first and send that , but then people would complain about the delay.

Anonymous 0 Comments

because even a wrong answer sounds more convincing when its convoluted with extra human sounding words.

If you don’t understand the explanation, you’re less likely to question it.

#42

Anonymous 0 Comments

Most of the tools you will be using (generative ai) will be building their response, next word, next pixel, next frame, based on the probability of the next word, pixel, frame being correct. That is why you are seeing it come in word by word. Equally they could build the entire sentence for you and send it at once, but this is more engaging and magical 🙂

Anonymous 0 Comments

It doesn’t do that because it can’t, that’s not not how it works. It doesn’t form complete thoughts and express them, it comes up with the most plausible word it can think of to go next after what it’s already written. It is just a more advanced version of  the autocomplete function on your phones keyboard. That’s why it’s so dangerous to use it for research, it will give you fake answers if they sound convincing enough because it doesn’t know the difference

Anonymous 0 Comments

This is a pretty simple answer. LLMs like ChatGPT do not have fully formed thoughts, they generate what you read word for word based on it’s training data.

It literally has no idea what word will come next until the previous word has been output.

Now they could, if they wanted, program in a delay, where the server buffers up all the words and prints them all at once. But that is extra complexity and server overhead that nobody wants to pay for.

Anonymous 0 Comments

Imagine ChatGPT is like a chef making your favorite meal. Instead of serving the whole dish at once, it carefully adds one ingredient at a time to make sure everything tastes just right.

Similarly, ChatGPT puts together words bit by bit to make sure they’re easy to understand and on point. This method helps avoid confusion and ensures the answer fits the question perfectly, like putting together a puzzle piece by piece.

Anonymous 0 Comments

When someone asks you a question think about the process of how you answer. YOu start saying words one at a time and your brain gives you each word one by one as you say them. Chatgpt is doing sort of the same thing. Each word is given and it then decides what the next word should be and so on and so on