What are the different types of AI?

Terms used:

The text you provide to a language model is called a prompt.

The result produced by the model is known as a completion.

The process of using the model to generate text is called inference.

LLMs use vector embeddings to represent textual elements, with each word mapped to a token ID and transformed into a vector. These mathematical representations enable efficient storage and searching of data, as well as identification of semantically related text.

A vector database is important because it enables efficient storage, retrieval, and manipulation of vector embeddings. By assigning a unique key to each vector, the database allows for quick and direct access to the content at a discrete level. This capability is par ticularly valuable in applications like RAG, where rapid retrieval and matching of vectors allow the model to discover semantically related text, such as a product that’s similar to one that a cus tomer searched for previously.

Vector Embedding:

Semantic caching

Feature Injection:

Feature injection refers to the process of incorporating additional information or features into the AI model. Although feature engineering is a general practice in data science, feature injection is a specialized technique used to enhance the performance andreasoning capabilities of AI models. Features can improve the model’s ability to handle specific tasks. By injecting relevant features, the model can capture and leverage important patterns in the data, leading to improved performance.


By introducing features that are relevant to a specific prompt, LLMs can gain a deeper understanding of the data and can better capture complex patterns and relationships.

Context retrieval involves retrieving relevant contextual information to enhance the understanding of AI models. By considering the surrounding context, such as previous interactions or user history, AI systems can generate more accurate and personalized completions. For example, a customer support system might use context retrieval to provide personal assistance. If a customer has previously interacted with the support system and mentioned a specific issue or order number, the system can retrieve that context to better understand the customer’s current inquiry or concern. Retrieval-augmented generation (RAG), a type of in context learning (ICL), is also important in this context. See Chapter3 for additional information.

Context retrieval

Among LLMOps tools, orchestration frameworks can be used to coordinate AI agents and other components to accomplish specific goals. AI agents are simply individual instances of language models that are responsible for performing specific tasks, such as text summarization, language translation, and sentiment analysis. They’re coordinated and managed within an orchestration system to complete complex language processing tasks. This process, known as orchestration, involves organizing agents, coordinating the input/output of various models, and managing the flow of data and information among agents. For example, in an e-commerce scenario, a chatbot interacting with a customer might use AI agents to retrieve order details from a database, generate a request for a return label using a shipping partner’s API, confirm the customer’s information, and initiate the return process by sending a shipping label.

Orchestration: