An Introduction to AI Agents

Learn the basics of AI agents, focusing on simple reflex agents that act on current perceptions with sensors, actuators, and rule sets. Explore strengths, limits, and a Python tic-tac-toe example.

Explain model based reflex agents that use an internal world model and condition action rules to act in partially observable environments, with self-driving cars and Pac-Man as examples.

Explore goal based agents that use an environment, a knowledge base, and search algorithms to plan actions and pursue long term goals, with examples like treasure hunt and chess.

Explain utility based agents and how they maximize expected utility through a utility function to guide actions. Include Python implementation using rock paper scissors and discuss advantages, limitations, and applications.

Learn how learning agents improve performance through experience, adapt to environments, and generate new knowledge. Explore supervised, unsupervised, and reinforcement learning with a tic-tac-toe Q-learning example.

Learn how hierarchical agents decompose problems into subagents, coordinate across levels, and learn via reinforcement learning and q-learning to optimize goals in robotics, gaming, and natural language processing.

Explore the Langroid Agent Library in this lecture—a multi-agent framework for language model apps that works with any model, enabling function calling, multi-prompt interactions, and retrieval augmented generation.

LaVague demonstrates an open-source, agent-based system that interacts with websites via a single function, enabling modularization of functions, using GPT-3.5 and OpenAI's API, and Huggingface takes notice.

Discover how GPT LM trainer enables easy fine-tuning via Colab, using API keys, prompts, and generated data to train multilingual models and assess translation tasks.

Explore how lm model frameworks break API calls into planner, executor, and summarizer functions within a swarm of agents, using environment concepts to fetch facts and orchestrate tasks.

Question the idea of using language models for everything by designing cost-aware ETL workflows for PDFs, CSVs, and databases that minimize language model usage with rules-based and ML options.

Plan and orchestrate AI-driven RFP workflows using Llama Index tools and rag tuning to parse proposals, provide context, build vector stores, and generate responses with a multi-agent framework.

Explore the basics of word vectorization and vectors, from two-dimensional and three-dimensional visuals to multi-dimensional concepts and complex numbers, and how vectorization represents words and sentences in language models.

Explore swarm algorithms paired with language model brains to perform API calls and function calls, with dynamic agents that communicate and produce structured JSON outputs.

Explore smolagents, a lightweight open-source agent platform with a toolbox of tools and DuckDuckGo search, default llm support, and a focus on supervision and reliability.