From the course: Azure OpenAI: Generative AI Models and How to Use Them
What are embeddings? - Azure OpenAI Service Tutorial
From the course: Azure OpenAI: Generative AI Models and How to Use Them
What are embeddings?
- [Instructor] What are embeddings? Embeddings are the numerical representation of words, sentences or other textual elements. They capture the semantic meaning, and the relationship between different elements in a dimensional space. Well, that's quite a difficult explanation, right? But let's look into what the problem is. The biggest problem is that, computers don't understand the meaning of text. That means when you want to try machine learning models based on text, you somehow need to convert them into numbers. 'Cuz again, computers do not understand text. They don't understand the meaning of text to be more specific. So just sending text to a machine learning algorithm or a computer will not do it. So what we do need to do is convert to texts into numbers. And once that is done, we can send it to our computer or to our machine learning algorithm. And that is actually what embedding models do. They will convert the text into a vector of numbers, but it's a bit more than just numbers. It's relationships. 'Cuz if you think about the relationship between the word cow and the word bull, we will know that they are quite similar, and that these words belong to each other. But if you think about the relationship of the words cow and chicken, although they're both animals, the distance between those numbers that needs to be generated will be farther away from each other than between a bull and a cow. So these models do not just convert text into numbers. The numbers actually mean something. The models know the relationship between words when they belong to each other or don't belong at all together. These models have been trained on top of textual data of many different kinds of sources. That way it learns the relationship between words and sentences. 'Cuz of that, that is possible to figure out what documents are similar. Or we can use this to search through documents based on natural language. So to repeat myself. Embedding models convert your text and to a vector of numbers. But those numbers contain the semantic meaning of your sentences. It's the only way how a computer can understand what text actually means.
Practice while you learn with exercise files
Download the files the instructor uses to teach the course. Follow along and learn by watching, listening and practicing.