Learn how large language models work

14 in-depth guides, about 33K words in all, written in plain English and checked against primary sources. Start with the fundamentals or jump to what you're building.

Updated October 10, 2026GlossaryVideos

Fundamentals

FundamentalsContext windows explainedA context window is the token budget a model can read and write in one request. This guide converts tokens to words and pages, compares current limits and explains the cost, quality and memory trade-offs of long prompts.Beginner11 min read FundamentalsHow large language models work, step by stepFollow one sentence through a large language model, from tokens and embeddings through self-attention, logits and sampling, with small worked examples at each step. Then see how training, scaling and mixture-of-experts shape what models can do, and why they hallucinate.Intermediate11 min read FundamentalsWhat is a large language model?A large language model is a neural network trained on enormous amounts of text to predict what comes next, which turns out to be enough to draft, summarize, translate and write code. This guide explains what LLMs are, how they're trained, the kinds you'll meet, the major families as of October 2026 and the limits to plan around.Beginner10 min read

Choosing & buying

Choosing & buyingHow to choose an LLM for your projectA practical framework for choosing a large language model: define the job, weigh ten criteria in a scoring matrix, and run a quick evaluation on your own data. Includes starting points for five common use cases as of October 2026 and a procurement checklist.Intermediate10 min read Choosing & buyingLLM API pricing explainedLLM APIs bill per token, but output, reasoning, caching, batching and long-context tiers can change a bill several-fold. This guide explains each pricing component, shows real prices as of October 2026 and works through three cost examples step by step.Beginner11 min read Choosing & buyingOpen-source vs closed LLMs: how to decideOpen-weight models let you run an LLM on your own infrastructure, but 'open' covers licenses ranging from Apache 2.0 to custom terms with user caps and naming rules. This guide compares open and closed models on quality, cost, control and risk, works through a total cost of ownership example and shows when a hybrid setup wins.Intermediate9 min read

Building

BuildingAI agents explainedHow AI agents work under the hood: a language model calling tools in a loop with memory, the planning patterns they use, and where the Model Context Protocol fits as of October 2026. Covers frameworks, evaluation, cost, security and the tasks agents handle well today.Intermediate11 min read BuildingFine-tuning vs RAG vs prompting: which should you use?Prompting changes the instructions, RAG changes the knowledge a model can see and fine-tuning changes its behavior. This guide gives a decision tree, a cost and risk comparison, fine-tuning methods, managed options as of October 2026 and worked examples.Intermediate9 min read BuildingPrompt engineering: a practical guide with examplesA practical guide to writing prompts that work: the five parts of a prompt, core principles, eight techniques with before-and-after examples, and how reasoning models change the rules. Includes a testing workflow and a reusable template.Beginner11 min read BuildingRetrieval-augmented generation (RAG) explainedA practical guide to retrieval-augmented generation, covering each stage of the pipeline from parsing and chunking to hybrid search, reranking and cited answers. It also explains how to evaluate RAG quality, when long context or fine-tuning fits better and how to secure the system.Intermediate10 min read

Running & deploying

Running & deployingHow to run LLMs locallyA practical guide to running open-weight language models on your own computer, from memory math and quantization formats to tools and model picks. It includes an Ollama quick start, hardware tiers as of October 2026, performance tips and fixes for common errors.Intermediate11 min read

Safety & quality

Safety & qualityLLM benchmarks explainedWhat the major LLM benchmarks test, which ones still separate frontier models as of October 2026, and why arena rankings and launch charts need a skeptical read. Includes a method for building your own evaluation with golden sets, rubrics and cost metrics.Intermediate9 min read Safety & qualityLLM hallucinations: why they happen and how to reduce themHallucinations are fluent, confident outputs that are false or unsupported by the sources a model was given. This guide explains why they happen, what published measurements show, and how to reduce, catch and measure them in real applications.Beginner10 min read Safety & qualityLLM security and privacy: a practical guideHow LLM applications get attacked and how they leak data, from prompt injection to over-privileged agents, with defenses that work in layers. Covers the OWASP Top 10 for LLM applications, vendor data policies, compliance basics and a pre-launch checklist.Intermediate11 min read

Building with LLMs? Get a vendor-neutral plan.

Tell us what you're building. Within 2 business days you get a model shortlist, an architecture outline and a cost estimate, plus optional introductions to vetted implementation partners.

Free and vendor-neutral: no model provider pays for placement.
Your details go only to partners you agree to.