Stage 01 of 11 · LLM Foundations · 5 min · Reviewed Aug 2026
LLM Foundations: History, Providers, Context, and Reasoning
Before agents, tools, or RAG, you need a working model of the model: what it is, what it is not, how providers differ in 2026, and how context windows, caching, and extended thinking change the bill. This is the floor the rest of the curriculum stands on.
- History
- Providers
- Model selection
- Context
- Reasoning
What you learn
How LLMs work and which model to choose for a given job, budget, and risk.
What an LLM is
A large language model predicts the next token given a context. That is the whole mechanism. Fluency, tools, and “reasoning” are all arrangements of that loop. The model does not retrieve a private database of facts unless you attach one. It does not see a URL, a file, or a browser unless a tool returns text.
Treat the model as a programmable text engine with a probability distribution — not a person, not a search engine, not a compiler. Agents, RAG, and workflows are software you wrap around that engine.
Visual
Next-Token Prediction Pipeline
An LLM receives input tokens, processes them through transformer layers, produces a probability distribution over the vocabulary, and samples the next token. This loop repeats until a stop condition is met.
A short history that still matters
The public story runs decoder-only transformers (GPT), instruction tuning and RLHF, then tool use (function calling), then long context, then test-time compute (reasoning / extended thinking). Each step changed the product, not just the benchmark.
What you inherit in 2026: chat APIs with system and user roles, structured outputs, tool calling, prompt caching, and optional thinking budgets. If you skip that stack and only paste a persona, you are using a 2022 interface on a 2026 model.
{
"model": "fast-extract",
"messages": [
{ "role": "system", "content": "Extract fields. Do not chat." },
{ "role": "user", "content": "Ticket #4412: laptop will not boot." }
],
"tools": [],
"thinking": { "budget_tokens": 0 }
}