Fact Zap

How do AI chatbots like ChatGPT work ?

ChatGPT is a chatbot. Chatbots are computer programs that can chat like humans. If you give them a prompt word or sentence, they start a conversation about the prompt. They can reply back, answer any question, search the internet and summarise their findings. Explore more on AI powered chatbots.

Artificial Intelligence illustration

Mobile phones give word suggestions when you type a message. This auto complete feature on mobile phones is an early version of AI (Artificial Intelligence). It looks at words as tokens. Sentences in a message are split into separate words. Every word is mapped with the word that comes before and after it. This gives AI a list of words that may come before and after every word in the sentence. As time passes, the mobile phone has a big mapping of related words.

After this training phase, the program is ready for real life testing. If we type in a word, it can give us a list of words that have high probability of coming after this word. If you have already typed in 3 or 4 words, it can give a highly accurate suggestion for the next word as it uses the 4 word pattern to search through its mapping. Complex mathematical equations (using probability theorems) are used by these programs to make the word mapping. Programs that can process words in a language and sequence them are called Language models.

Large Language Models

Large Language Models can map sentences and even paragraphs. They are so complex that they make many layers of mappings. They store sequences and probabilities for billions of word combinations.
If they are asked a question like "How long does it take to travel to Mars", Large Language models process the question to find focus words like MARS, HOW LONG, etc. It will come up with sentences that have high probability for the focus words. Using these sentences and their probabilities, it can give you the answer in one sentence that looks a lot like sentences written by humans. It can join sentences and come up with an essay on the focus works. They need massive volumes of data for training.


Large Language Models are a type of Artificial Intelligence Programs. There are similar models that can process sounds, shapes, pictures, etc. Language models come up with words and sentences. Other AI libraries and models can make sounds, images, art, etc that look a lot like the ones made by humans.
Chatbots without AI are trained on a given dataset. Their knowledge of words and subjects are limited.
Chatbots powered by AI learn on the go. Everytime they search the internet, they add more information to existing mappings. Their datasets keep increasing with time.


ChatGPT is an application. It is available as a website and as a mobile app. AI chatbots like ChatGPT do not really understand where Mars is located, or what it means by space travel. You can download it or use the website version. Inside ChatGPT, users can type a word, sentence or question. Say, I ask ChatGPT about space travel to Mars. It answers back with sentences and essays on space travel to Mars and ongoing space missions. This information is from existing websites and online sources.
In short, ChatGPT uses information available online to mimic human conversations.

Can AI pose a threat to humans ?

As of now, AI programs are made by humans, trained on human made data, programmed to produce outputs that resemble the ones made by humans. They are designed to take over repetative tasks. They make life easy for humans. They cannot discover or invent. If used well, AI can help humans think, create and progress more.

AI bots are designed to make life easy for humans. They cannot work without humans and human made information.