Language model learning hub
Transformers and Large Language Models
Large language models generate text through an autoregressive loop built on transformer representations. Tokens become vectors, attention moves information between allowed positions, the model scores the next token, and a decoding policy selects a continuation. Reliability still depends on evaluation, grounding, and application controls.
Best for: Learners who want a mechanism-first path from tokens to production LLM behavior
Recommended learning order
- 1How LLMs Generate TextSee how a language model predicts, selects, and appends one token at a time.Open lesson
- 2LLM TokenizationUnderstand how text becomes versioned token IDs, context usage, and model cost.Open lesson
- 3TransformersThe attention mechanism that powers modern AI.Open lesson
- 4Embeddings & Context WindowsLearn how models represent text and how finite context changes application design.Open lesson
- 5LLM Decoding & SamplingCompare greedy decoding, temperature, and top-p through visible probability changes.Open lesson
- 6LLMsLarge Language Models and the GPT architecture.Open lesson
- 7LLM Limitations & GroundingRecognize unsupported output and design evidence, citations, abstention, and evaluation.Open lesson
Decisions this path should help you make
- How do tokenization choices change cost and behavior?
- What information does attention route between token positions?
- When should decoding be deterministic or stochastic?
- Which claims require external evidence rather than model confidence?