AI Basic Concepts

Artificial Intelligence (AI)
Artificial Intelligence is the field of building machines that mimic human intelligence. These systems learn from data, solve problems, and make decisions without explicit programming. AI powers applications like self-driving cars, virtual assistants, and fraud detection by adapting to real-world scenarios.
Example: AI enables Tesla’s autopilot to navigate and respond to traffic.

Machine Learning (ML)
Machine Learning is a core subset of AI that enables systems to learn patterns from data and improve over time. ML models can detect trends, classify inputs, and make predictions without being explicitly programmed for each task. It's foundational to AI advancements in almost every industry.
Example: Netflix suggests movies based on your previous viewing habits.

Deep Learning
Deep Learning is a type of machine learning that uses neural networks with many layers to analyze complex data. It’s especially powerful for tasks involving images, speech, or language due to its ability to model high-level abstractions. It requires large datasets and computational power to perform well.
Example: Google Photos can detect and group people by recognising faces.

Natural Language Processing (NLP)
NLP helps machines understand and generate human language. It combines linguistics with machine learning to interpret text, speech, or commands. NLP is used for chatbots, sentiment analysis, language translation, and more.
Example: ChatGPT holds realistic conversations or answers complex queries in natural language.

Computer Vision
Computer Vision allows machines to interpret and understand visual inputs like photos or videos. It enables applications such as object detection, facial recognition, and visual search. By processing pixels and patterns, it helps AI "see" the world.
Example: Your iPhone unlocks using Face ID through computer vision.

Supervised Learning
Supervised Learning trains models using labeled data, where the correct outputs are already known. The model learns to map inputs to outputs and generalise to new examples. It's widely used in email filtering, diagnostics, and classification problems.
Example: Spam detection systems are trained with labeled spam and non-spam emails.

Unsupervised Learning
Unsupervised Learning involves analysing data without predefined labels to find hidden patterns or structures. It’s used for clustering, anomaly detection, and dimensionality reduction. This technique helps discover insights from raw data.
Example: E-commerce platforms group similar customers based on shopping behavior.

Reinforcement Learning
Reinforcement Learning teaches an agent to learn optimal behaviour through rewards and penalties. It makes decisions by exploring, acting, and adjusting over time to maximise long-term success. It's ideal for sequential decision-making problems.
Example: AlphaGo mastered the game of Go by playing millions of simulations and learning strategy.

Neural Networks
Neural Networks are the backbone of deep learning, designed to mimic how the human brain processes information. They consist of layers of nodes (neurons) that learn from data through weight adjustments. They excel in recognising patterns and making predictions.
Example: OCR systems use neural networks to read handwritten postal codes.

Generative AI
Generative AI creates new content—text, images, music, or code—based on what it has learned. It uses models like GANs or transformers to produce original, high-quality content. Generative AI is revolutionising creativity, design, and automation.
Example: DALL·E generates images from descriptive prompts like "a futuristic city on Mars."

Large Language Models (LLMs)
LLMs are powerful models trained on massive text datasets to understand and generate human-like language. Built on transformer architecture, they can perform tasks like summarisation, question answering, and translation. LLMs understand context, tone, and meaning.
Example: GPT-4 writes emails, answers questions, or summarises legal documents.

Popular LLMs by Organisation
OpenAI: GPT-4
Meta: LLaMA
Google: Gemini
Microsoft: Phi-2
Anthropic: Claude 3.5
Mistral AI: Mistral
xAI: Grok

Closed vs. Open LLMs
Closed LLMs are proprietary, with limited access to source code or training data.
Example: GPT-4 (OpenAI), Grok (xAI)
Open LLMs are open-source and allow community use, modification, and research.
Example: Meta’s LLaMA series is widely used in academia and startups.

Transfer Learning
Transfer Learning involves taking a model trained on one task and adapting it to a new, related task. This allows models to be reused with less data and compute. It boosts performance in specialised domains using general pre-trained models.
Example: Adapting an ImageNet-trained model to detect lung cancer in X-rays.

Fine-Tuning
Fine-tuning refines a pre-trained model using a smaller, domain-specific dataset to improve performance. It adjusts the model’s parameters so it understands the unique features of the new task.
Example: Fine-tuning GPT on legal documents to turn it into a legal assistant.

Types of Fine-Tuning
Parametric Fine-Tuning updates the internal weights of a model to adapt it to a new task.
Example: Retraining GPT on customer feedback for improved sentiment analysis.
Non-Parametric Fine-Tuning adds external tools—like memory or retrieval—without altering the model’s core.
Example: Using RAG to provide external context to GPT without changing its base model.

Prompt Engineering
Prompt Engineering is the practice of designing effective inputs to guide LLMs to desired outputs. Clear, structured prompts improve model accuracy and control. It’s a vital skill in low-code and no-code AI workflows.
Example: Asking “Summarise this article in 3 points” gives a concise output.

Prompts and Their Types
Simple Prompts are direct and factual.
Example: “What is the capital of Japan?”
Contextual Prompts add background or role.
Example: “As a food critic, describe Tokyo’s cuisine.”
Instructional Prompts guide specific tasks.
Example: “Translate this text into French.”
Multi-turn Prompts involve interactive, back-and-forth dialogue.
Example: A tutoring session that builds on previous questions.

Vision-Language Models (VLMs)
VLMs integrate text and image understanding to enable multimodal AI. They generate, describe, or match visual content using text inputs. They enable deeper context in image analysis and creation.
Example: CLIP can match a prompt like “a dog with sunglasses” to the correct image.

Retrieval-Augmented Generation (RAG)
RAG enhances LLM performance by retrieving relevant documents from an external knowledge base before generating answers. It improves accuracy, relevance, and real-time awareness.
Example: A customer support bot retrieves internal policy docs before replying.

Vector Databases
Vector databases store and retrieve data based on vector similarity, making them ideal for semantic search. They are essential for storing embeddings from AI models and performing nearest-neighbor queries.
Example: Searching for “sunset beach” retrieves visually similar images, even without exact keywords.

LangChain Ecosystem
LangChain is a framework for building advanced LLM applications through modular workflows. It connects components like memory, tools, and data for complex AI behaviour. LangServe handles backend requests, LangSmith supports testing, and LangFlow offers a visual interface.
Example: A contract analysis tool that extracts clauses and fetches definitions using LangChain.

Hugging Face Ecosystem
Hugging Face is an open-source AI platform that provides tools, models, datasets, and libraries to build, train, deploy, and share machine learning models—especially those focused on NLP and LLMs. It has become the go-to hub for transformer-based models, collaboration, and research in modern AI.
Example: Developers use Hugging Face to quickly deploy a sentiment analysis model trained on tweets.

AI Agents
AI Agents are autonomous systems that act on instructions and make decisions. Unlike simple chatbots, they can plan tasks, invoke tools, and interact with software systems. They can be used in automation, research, or operations.
Example: An AI agent that reads your emails, adds meetings to your calendar, and books tickets.

Multi-Agent Systems (MAS)
Multi-Agent Systems involve multiple intelligent agents collaborating toward a shared goal. Each agent has specific capabilities, but together they solve complex problems. These systems simulate human teams in distributed environments.
Example: In logistics, one AI handles routing, another inventory, and a third customer notifications.

Explainable AI (XAI)
XAI aims to make AI systems transparent, interpretable, and trustworthy. It allows users to understand why a decision was made—critical for regulation and accountability.
Example: A diagnostic AI showing which symptoms led to a cancer diagnosis.

Diffusion Models
Diffusion models generate content by gradually refining noise into detailed outputs. They are particularly good at generating high-quality, realistic images.
Example: Stable Diffusion turns a text prompt like “a cyberpunk city at night” into stunning digital artwork.

AI Agents vs. Chatbots
Chatbots are primarily conversational and reactive. AI Agents, on the other hand, can take initiative, perform tasks, and interact with multiple tools. Agents are proactive and goal-driven.
Example: A chatbot might answer a question, but an AI agent can book a flight based on a travel query.