AI and large language models

<p>Learn the computer science foundations you need to jump start your career. Ten chapters covering all the career-essential topics. Become confident in computer architecture, networking, databases and much more. Perfect for beginners, self-taught developers, bootcamp graduates and anyone who wants to really understand how computers work.</p>

Introduction

How large language models work

From text to tokens

Why tokenise at all?

How tokenisation works

The consequences of tokenisation

The transformer architecture (revisited)

The problem attention solves

How attention actually works

Causal masking

Multi-head attention

Stacking layers

The complete transformer block

Training: predicting the next token

The training objective

How learning happens

Self-supervised learning

Why scale matters

Emergent capabilities

The training data

The cost of training

Alignment: from prediction to assistance

Supervised fine-tuning

Reinforcement learning from human feedback

Why RLHF works

Constitutional AI and alternatives

The result of alignment

Fine-tuning: customising models

When to fine-tune

How fine-tuning works

Full fine-tuning vs. efficient methods

Practical fine-tuning considerations

The context window

Managing limited context

Lost in the middle

What LLMs are (and aren’t)

What they’re good at

Why they hallucinate

Why they’re inconsistent

Why they struggle with math

Reasoning vs. pattern matching

Why knowledge has limits

Why coherence degrades

The understanding question

What’s happening inside?

Probing for concepts

Attention patterns

Circuits and features

Why interpretability matters

Working with LLMs

The API paradigm

Prompt engineering

The system prompt

Principles of effective prompts

Few-shot prompting

Chain-of-thought prompting

Why small changes matter

Structured outputs

Tool use and agents

How tool use works

Agents

Planning and decomposition

Building LLM applications

Retrieval-Augmented Generation (RAG)

RAG architecture details

RAG evaluation

How embeddings work

Chunking strategies

Building conversational interfaces

Debugging LLM applications

Content generation and transformation

Security considerations

Cost optimisation

Evaluation and testing

Monitoring and observability

Benchmarking and model selection

Public benchmarks

Custom evaluation

Model providers comparison

The evolving landscape

Multimodal models

Vision-language models

Image generation

Audio models

Multimodal integration

Safety and responsible AI

Misuse risks

Bias and fairness

Privacy concerns

Environmental impact

Regulatory landscape

Smaller, specialised models

Running models locally

Model serving at scale

Inference costs and latency

Quantisation for deployment

Scaling infrastructure

The API vs. open-source decision

Beyond language: world models

Architecture variations and alternatives

Staying current

The broader context

Conclusion