AI Agents and Compound AI Systems

Flashcards on AI Agents and Compound AI Systems

Limitations of Monolithic Models

Limited by the data they've been trained on, impacting their knowledge and task-solving abilities; hard to adapt, requiring data and resource investment for tuning.

Compound AI Systems

AI systems that integrate models into existing processes, allowing them to access databases and other external resources for more accurate and comprehensive problem-solving.

System Approach

The ability to break down a desired program function and select the appropriate components to achieve it, offering a faster and more adaptable solution than tuning a model.

Retrieval Augmented Generation (RAG)

A popular Compound AI system where models are prompted to create a search query that can go into a database, fetch information, output an answer, and then, that would go back into the model that can generate a sentence to answer.

Control Logic of a Program

The defined path to answer a query in a Compound AI system, often programmatically controlled by a human.

Agentic Approach

A Compound AI system where a large language model (LLM) is in charge of the control logic, leveraging the LLM's reasoning capabilities to solve complex problems.

Reasoning

The capability of LLM agents to develop a plan and reason about each step of the process, putting the model at the core of problem-solving.

Tools

External programs that LLM agents can call to execute solutions, such as search engines, databases, calculators, or other language models.

Memory

The ability of LLM agents to store and retrieve inner logs or conversation history for more personalized and context-aware interactions.

React

A configuration approach for AI agents that combines the reasoning and acting components of LLMs.

Sliding Scale of LLM Autonomy

A system that examines the trade-offs and what they want in terms of an autonomy of the system.