The Concepts
Let's first define some key concepts.
4 min read
Before jumping into details, let's first define some key concepts (in alphabetical order).
AI Agent
An artificial intelligence (AI) agent is a software program designed to perform tasks autonomously, without direct human intervention.
AI agents can interact with their environment, gather information, and take actions to achieve specific goals. They are specialized for various purposes, including task automation, decision-making, problem-solving, and natural language interaction, often with the ability to adapt or improve over time.
Source(s):
(https://startup-house.com/glossary/ai-agent-definition)
Generative AI
Generative AI is a branch of artificial intelligence focused on building models that can create new content by learning patterns from training data. This content can include text, images, audio, or other forms of data.
In probabilistic terms, generative AI aims to learn the probability distribution p(x) over a dataset X, representing the likelihood of observing a specific data point x. The ultimate goal is to generate new samples that align with this learned distribution. Generative models attempt to capture the underlying structure of the data by estimating these probabilities.
In contrast, discriminative models focus on estimating the probability of a label y given an observation x, or p(y|x).
Bridging both concepts, conditional generative models seek to model the conditional probability of observing x given a label y, i.e., p(x|y).
Source(s):
David Foster. Generative Deep Learning, 2nd Edition. O'Reilly Media, 2023
LLM
A Large Language Model (LLM) is a type of artificial intelligence that can recognize and generate text, among other tasks. LLMs are trained on large datasets, often consisting of billions of words, and are built using neural networks, particularly transformer models.
A more technical definition describes an LLM as a neural network with billions of parameters, trained on massive quantities of unlabeled text using self-supervised learning techniques.
Source(s):
Cloudflare
Graphable
The paper A Comprehensive Overview of Large Language Models provides a detailed exploration of LLMs, focusing on their architecture, particularly transformer models and attention mechanisms. It covers key concepts such as tokenization, positional encoding, and distributed training. You can access the full paper here:
RAG
Retrieval-Augmented Generation (RAG) is an AI architecture that enhances large language models (LLMs) by incorporating external information to improve the accuracy and relevance of their outputs. It retrieves relevant data from sources like databases or documents, using it to provide additional context for the model. This retrieved information is combined with the original query in the prompt, enabling the model to generate responses grounded in external knowledge.
Source(s):
Advanced RAG Techniques: an Illustrated Overview
Enjoyed this post? Found it helpful? Feel free to leave a comment below to share your thoughts or ask questions. A GitHub account is required to join the discussion.
