Replace LLM article with concise introduction

articles/llms-101.md

@@ -0,0 +1,36 @@
+# LLMs 101: What Even Is It?
+---
+### Preface
+
+Large language models (LLMs) are artificial neural networks trained on vast amounts of text. The most common type is the **transformer**, introduced in the 2017 paper _Attention Is All You Need_. Transformers break text into smaller units called **tokens**, turn those into numbers, run them through layers of matrix math and attention, and then predict what tokens come next.
+
+OpenAI’s ChatGPT, built on GPT-3, brought this architecture into the spotlight and kicked off today’s wave of AI applications.
+
+---
+### What can LLMs do?
+
+Although they’re called _language_ models, LLMs can handle many kinds of data if trained for it. For text, typical uses include:
+
+- Generating text    
+- Completing text
+- Translating text
+- Analyzing or summarizing text
+- Answering questions
+- Powering agents that perform tasks
+
+This Cookbook focuses on the first few, since they’re the foundation of most real-world applications.
+
+---
+### How do they work?
+
+At their core, LLMs are **prediction engines**. Given some input text, they predict the most likely next token, then the next, and so on until they produce a coherent output.
+
+Tokens live in an **embedding space**—a mathematical map where related concepts sit near one another. For example, _king_ and _queen_ are neighbors, just as _emperor_ and _empress_ are. This structure allows the model to capture relationships and context when generating text.
+
+---
+### **Why it matters for you**
+
+You don’t need every detail of the math to use LLMs effectively. The key takeaway is simple:
+**LLMs generate outputs by predicting one token at a time.**
+
+The rest of this Cookbook will show you how to harness that prediction power—whether you want to generate text, complete thoughts, or analyze documents.