Dear children, hello!
Today we are going to talk about our good friend - ChatGPT. ChatGPT is very popular recently, and you may have seen articles and videos about it in many places.
You may ask, what exactly is this ChatGPT? seems very amazing! How does it work?
Don't worry, I will take everyone to uncover its mystery next.
The principle of ChatGPT
First, let's get to know ChatGPT. It is an artificial intelligence chatbot developed by OpenAI. Its full name is Chatbot Generative Pre-training Transformer, or ChatGPT for short. The name may seem complicated, but how it works is actually not that difficult to understand.
At the heart of ChatGPT is something called a language model. This language model is like a super smart student. It has learned human language rules and knowledge by reading a lot of books and articles. The process is like learning languages when we were kids, only it learns much faster than we do—faster than you might think.
Language models work on the basis of probability. Just like we are playing a word guessing game, it will predict what the next word will be based on probability, and this prediction is based on the previous words. For example, if I say "I love to eat ____", assuming that the next word may be "apple" with a probability of 50%, "chocolate" with a probability of 30%, and "pizza" with a probability of 20%, then it Will answer "I love apples".
The language model predicts the next most likely word based on what it has learned.
ChatGPT uses a special language model called Transformer. The name sounds like a transformer, but it's actually a very powerful language model. It can handle long sentences, understand complex language structures, and even generate coherent text.
The working principle of Transformer is through the self-attention mechanism. This mechanism is like a super smart detective, it can pay attention to every word in the sentence and understand the relationship between them. For example, if I say "I put the book on the table," then the detective knows that "I" is the subject, "book" is the object, and "table" is the location.
The training process of ChatGPT is like a big fill-in-the-blank game. It looks at sentences with gaps and tries to fill in the best words. The process is like the fill-in-the-blank games we played as kids, only this time on a much larger scale.
Do you remember that "fill in the blank" game we played when we were kids?
For example, "I ate a big ____", you might immediately think "apple" or "watermelon", right?
For example, "the earth is ____", you might immediately think "round", right?
This is because our brains have learned language models. And now, there is a kind of artificial intelligence, which is like a super "fill in the blank" player, which is the language model of artificial intelligence.
This super player, its scale may be beyond our imagination. For example, "the earth is ____", it will not only answer "round", but also may answer "blue", "our home", "source of life", or even "two-thirds covered by water ".
This is the power of language models.
You can think of this power player as a tireless bookworm. It never needs to sleep, it never needs to rest, it just needs to constantly read various materials, such as all books, news, web pages, even our social media posts, which are its "food". After a long time of reading and studying, it begins to understand the relationship between words, just like we can understand and predict the plot of a story.
ChatGPT may speak gibberish
ChatGPT is such a super player. They are state-of-the-art language models that can understand complex sentences and even write articles of fairly high quality. But just like we sometimes misunderstand the plot of a story, language models can also make wrong predictions. They may misinterpret the meaning of the text, or generate sentences that do not correspond to reality.
For example, if we ask it "how's the weather today?", it might answer "the weather is fine today". But in fact, it doesn't know the real weather conditions today. It just makes the most likely answer based on similar sentences it has seen before.
For another example, if we ask it "Is the moon made of cheese?", it might answer "Yes". While this answer is grammatically correct, in reality the moon is clearly not made of cheese. That's because, while language models can understand the relationship between words and sentences, they don't understand the real world.
This will cause the language model to talk nonsense. It will only find the word with the highest probability in the model to answer our questions based on the probability. As for whether it is right or not, it does not know it.
Therefore, when we use ChatGPT, we can't be lazy and believe what it says 100%, we have to be careful to verify it ourselves.
Learning to ask questions is key
Overall, ChatGPT is like a combination of a super smart student and a detective. It learns human language rules and knowledge by reading a large amount of material, and then uses this knowledge to generate coherent text. While how it works may seem complicated, it's actually doing what we all did as kids: learn languages and fill in the blanks.
Because ChatGPT is "filling in the blanks" behind our questions, our questions are very important. The quality of the questions directly determines the quality of the answers. The words that continue to increase are like vines, all based on the previous sentence. grown out.
If the questions we ask are not good, it will lead to "the upper beam is not straight and the lower beam is crooked". This is why the same ChatGPT, some people think it is very smart and knows everything, but some people think it is stupid.
In order to ask good questions, there is a new profession called: Prompt Engineer, which is dedicated to helping others ask good questions to AI.
Now we know that the language model of artificial intelligence is like a super "fill in the blank" player, it does not create new information itself, all its information is "fed" to it when people train the language model, and then It's just broken, rearranged and combined output, so the underlying principle of ChatGPT is not complicated, not as mysterious as others say.
Uses of ChatGPT
Then you may ask, what is the use of this ChatGPT? What can it do for us?
It can do many things, it can help us answer questions, translate, write articles, and even write poems for us. It is like an all-round assistant, its function will not be limited to some special fields, no matter what you need, it can give you the answer, as for whether the answer is right or not, it still needs to be judged by you. It also has many interesting usages that you can explore by yourself. If you haven't used ChatGPT yet, try it out, I believe you will like it.
Well, that's all for today's sharing. I hope you enjoyed this article and I hope you learned something from it. You are welcome to forward this article to your friends, and you are welcome to leave a message in the comment area.
See you next time!