Artificial intelligence basics come down to a few nested ideas. AI is the broad goal of machines doing tasks that need intelligence. Machine learning is the main way we get there: systems that learn patterns from data. Deep learning is machine learning with many-layered neural networks. And large language models, the technology behind ChatGPT, Claude and Gemini, are deep learning models trained on huge amounts of text.
This guide explains each core concept in plain language, in the order they build on each other, with real examples. Want a quick glossary instead? See 20 basic AI terms explained simply.
How the Pieces Fit Together
| Concept | What it is | Everyday example |
|---|---|---|
| Artificial intelligence | The whole field: machines performing tasks that need intelligence | Everything below |
| Machine learning | Systems that learn from data instead of hand-written rules | Spam filters, recommendations |
| Deep learning | Machine learning with many-layered neural networks | Face unlock, speech recognition |
| Large language models | Deep learning models trained on text to predict the next word | ChatGPT, Claude, Gemini |
| Generative AI | Models that create new text, images, audio or code | Image generators, coding assistants |
Machine Learning
Traditional software follows rules a programmer writes. Machine learning flips that: you show the system many examples, and it works out the rules itself. A spam filter is not told “emails containing X are spam”; it learns what spam looks like from thousands of labeled emails.
There are three main ways a machine learning system learns:
Supervised learning
The model trains on labeled examples, inputs paired with the right answer, and learns to map one to the other. Examples: spam detection, predicting house prices, recognizing objects in photos, flagging suspicious transactions.
Unsupervised learning
The model gets unlabeled data and finds structure on its own, such as groups of similar items or unusual outliers. Examples: customer segmentation, anomaly detection in server logs, grouping similar documents.
Reinforcement learning
An agent learns by trial, error and reward: it tries actions, gets feedback, and adjusts to maximize reward over time. Examples: game-playing systems like AlphaGo, robot control, and the fine-tuning step that makes chat models more helpful (reinforcement learning from human feedback, or RLHF).
Transfer learning
Instead of training from scratch, you start from a model that already learned something general and adapt it to your task with much less data. Almost all practical AI today works this way: fine-tuning a pre-trained model is far cheaper than building one. Our guide to fine-tuning a local AI model walks through it hands-on.
Neural Networks and Deep Learning
A neural network is a stack of simple mathematical units (“neurons”) connected in layers. Each connection has a weight, and training adjusts those weights until the network’s outputs match the examples. Deep learning just means networks with many layers, which lets them learn complex patterns directly from raw data such as pixels, audio or text.
The main architectures you will hear about:
- Convolutional neural networks (CNNs): built for images and video.
- Recurrent neural networks (RNNs): an older design for sequences like text and audio, now largely replaced by transformers.
- Transformers: the architecture behind every modern language model. They process a whole sequence at once and use “attention” to decide which parts matter to each other.
- Diffusion models: the main technique behind today’s image and video generators. They learn to turn random noise into a picture step by step.
- GANs (generative adversarial networks): two networks, a generator and a critic, competing to produce realistic images. Important historically, now mostly overtaken by diffusion models.
Large Language Models (LLMs)
An LLM is a transformer trained on an enormous amount of text with one simple objective: predict the next token (a word or piece of a word). Scaled up far enough, that objective produces models that can write, summarize, translate, answer questions and write code. Examples include OpenAI’s GPT models, Anthropic’s Claude, Google’s Gemini, and open models like Llama, Qwen and Gemma that you can run on your own computer.
For a deeper but still practical explanation, including tokens, context windows and why models hallucinate, read What Is an LLM? A Working Model for Engineers. To run one yourself, see our LM Studio and Ollama setup guide.
Natural language processing (NLP)
NLP is the older, broader field of getting computers to work with human language: translation, sentiment analysis, speech recognition and summarization. LLMs are now the dominant technique within NLP, but the term still covers the whole area.
Prompts and prompt engineering
A prompt is the instruction you give a model. Clear, specific prompts with context and an example of the output you want get much better results than vague ones. “Summarize this report in 5 bullet points for an executive” beats “summarize this”.
Generative AI, RAG and AI Agents
Generative AI
Generative AI is any model that creates new content: text (LLMs), images and video (diffusion models), music, or code. It is the category that made AI mainstream.
Retrieval-augmented generation (RAG)
An LLM only knows what was in its training data. RAG fixes that by first searching your documents or database for relevant information, then passing it to the model along with the question. It is how most company chatbots answer questions about internal docs. Full explanation: What Is RAG?.
AI agents
An AI agent is an LLM that can take actions, not just reply: it plans steps, calls tools (search, code execution, APIs), looks at the results and keeps going until the task is done. Coding assistants that edit files and run tests are the most common example today. See AI Agents Explained: How They Actually Work.
AI Beyond Language
- Computer vision: understanding images and video. Used in face unlock, medical imaging, manufacturing inspection and self-driving cars.
- Speech AI: turning speech into text and text into natural-sounding speech.
- Robotics: combining vision, planning and control so machines can act in the physical world, from warehouse robots to surgical assistants.
Where to Go Next
If you want to understand AI as a developer, read our AI concepts series in order: What Is an LLM?, then What Is RAG?, then AI Agents Explained. If you prefer structured courses, Anthropic Academy’s free courses and our list of free AI courses are good places to start.
Frequently Asked Questions
AI is the broad goal of machines doing tasks that need intelligence. Machine learning is the main approach: systems that learn patterns from data instead of being explicitly programmed. Deep learning is machine learning using many-layered neural networks.
A neural network trained on very large amounts of text to predict the next token. That single objective is enough to make it generate coherent language, answer questions, and write code.
Supervised learning trains on labeled examples so the model maps inputs to known outputs. Unsupervised learning finds structure in unlabeled data, such as clustering similar items.
No. The core ideas can be understood conceptually. Linear algebra and statistics only matter once you build or fine-tune models yourself.