Introducing the first three foundational courses in the new Hextree AI continent.

Developing Intuitions for LLMs

At Hextree, we believe it is important to understand technology deeply, even if AI covers a lot of the actual work. And so we believe that developing an intuitive understanding of how LLMs actually work provides a lot of benefits when working with AI, as well as when attacking systems using AI.

LLMs are very complex and there are many layers to this technology. So in typical Hextree style, we start with the basics and build up to the complexity. These courses are perfect if you didn’t follow the evolution of LLMs and want to catch up.

We try to find a good balance between technical details and intuitive understanding. We are not diving deep into the actual math, but deep enough to get a sense of how LLMs can do what they do. We start with the common premise that LLMs are “just next token predictors”, briefly explore the history of AI, compare different LLM architectures, look at the role of the attention mechanism, and finally experience how agents actually work.

You can also experiment with our token visualizer to see how sampling and prompt templates affect the output. And in the agents course, we finally understand how function calls work.

Start Learning

Artificial IntelligenceHow LLMs Generate TextThis course explains the core mechanics and brief history behind modern LLMs. You'll learn about tokenization, embeddings, attention, sampling, special tokens and prompt templates, to develop an intuitive understanding of LLMs.
Artificial IntelligenceLLM Architecture and TrainingThis course builds a practical mental model of modern LLMs. You’ll understand how they are trained, how tokens move through transformer layers, different forms of attention, and what MoE, distillation, and quantization means.
Artificial IntelligenceIntroduction to AgentsLearn what actually happens when an AI assistant uses tools and completes a multi-step task. This course follows the process from structured LLM output and function calls to autonomous agent loops and code execution, giving you an intuitive understanding of modern AI agents.