๐ช๐ต๐ฎ๐ ๐๐ฎ๐ฝ๐ฝ๐ฒ๐ป๐ ๐ช๐ต๐ฒ๐ป ๐ฌ๐ผ๐ ๐ฃ๐ฟ๐ผ๐บ๐ฝ๐ ๐๐ต๐ฎ๐๐๐ฃ๐ง
You type a prompt. ChatGPT processes your words before it thinks.
Here are 4 steps.
Tokenization The AI chops your text into tokens. These are small pieces of words. Spaces matter. " ChatGPT" is a different token than "ChatGPT".
Vectorization AI models only understand numbers. The AI turns tokens into vectors. A vector is a list of 12,000 numbers. Similar words stay close. King and Queen are near. King and Apple are far. Language becomes math.
Transformer Layers Tokens move through 96 layers. Each token looks at other tokens. It assigns a weight. This helps the AI understand context. Example: The animal did not cross the street because it was tired. The AI knows "it" refers to the animal. No grammar books. Only math.
Sampling The AI predicts the next token. It uses probability. If you type "The capital of France is", Paris has a 95% probability. The AI picks the most likely word.
No reasoning happens here. No understanding happens here. The AI predicts the next word based on previous words. This process feels like thinking when scaled up.
Source: https://dev.to/saranyam_15/what-happens-when-we-type-a-prompt-into-chatgpt-3287