• Click to focus & highlight links
  • Double-click the node to visit note
  • Drag to rearrange
  • Click blank space to zoom out

Artificial Intelligence

Artificial Intelligence (AI), in its broadest technical sense, is the engineering discipline of creating systems capable of performing tasks that typically require human intelligence. These tasks include reasoning, learning, perception, problem-solving, and language understanding.

The Evolution: From Logic to Deep Learning

Historically, the field began with Symbolic AI (Good Old-Fashioned AI), which relied on explicit, hand-coded rules and logic. While effective for well-defined problems like chess, it was brittle in the messy real world. The paradigm shifted with Machine Learning, where systems learned rules from data rather than being explicitly programmed.

This evolution accelerated rapidly in the 2010s with Deep Learning. Companies like DeepMind (founded in 2010, acquired by Google in 2014) demonstrated that neural networks with many layers could master complex domains—from defeating Go champions (AlphaGo, 2016) to predicting protein structures (AlphaFold).

The Generative Era

The modern era of AI is defined by specific breakthroughs in Generative AI, driven largely by the Transformer architecture introduced in 2017 (“Attention Is All You Need”).

  • 2018-2020: The rise of BERT and GPT-3 showed that models pre-trained on vast internet corpora could perform few-shot learning.
  • 2022: The release of ChatGPT marked the “iPhone moment” for AI, democratizing access to Large Language Models (LLMs) via chat interfaces.
  • 2023-Present: The focus shifted to Multimodal models (understanding images, audio, and text simultaneously) and efficiency (Small Language Models).

The Agentic Future

The most recent frontier is AI Agents that could be implemetned and defined by Agent Skills. While a standard LLM waits for a prompt and generates text, an Agent is a system designed to perceive, reason, act, and reflect. Instead of just answering a question, an agent can break down a complex goal (“Plan a travel itinerary”) into steps, browse the web, book tickets, and adjust its plan if errors occur.

Capability Analysis

To deliver a modern AI architecture, we must look beyond just “the model” and consider the capability stack:

  • Foundation Models: The core reasoning engines (LLMs, Diffusion Models) that serve as the ‘brain’ of the system.
  • Vector Database: The long-term semantic memory that allows models to retrieve relevant information via RAG (Retrieval-Augmented Generation).
  • AI Engineering: The operational layer, including Prompt Engineering, fine-tuning, and evaluation frameworks (LLMOps) to make models reliable in production.
  • AI Safety and Alignment: The capability to ensure models remain helpful, honest, and harmless through techniques like RLHF (Reinforcement Learning from Human Feedback).
  • Autonomous Agents: The orchestration layer that enables tool use (function calling), planning loops (ReAct), and multi-agent collaboration to execute tasks, possibly through Agent Skills.

AI is no longer just a model in a notebook; it is a complex distributed system integrating Purposely-built Databases for memory and extensive tool integrations for action.

Generative AI Ecosystem

We can break down the modern Generative AI stack into first principles, understanding the role of each component in building intelligent systems.

At the core of this ecosystem is the Large Language Model, a probabilistic engine built on the Transformer architecture that processes sequences parallelly. The model reasons over text provided within its Context Window. This reasoning process is initiated and steered by a well-crafted Prompt, but raw models are limited by their training data. To bridge this gap, we employ RAG (Retrieval-Augmented Generation) coupled with Semantic Search to inject relevant, up-to-date facts directly into the context. Moving from observation to action, the model utilizes Function Calling to execute code or manipulate external tools, increasingly standardized via the Model Context Protocol (MCP). Finally, for complex, multi-step applications, orchestration frameworks like Langchain bind these components together, enabling the creation of autonomous systems where specialized SubAgent units handle distinct parts of a larger objective.

Updated: