Guide4 of 9 layers available
Everything required to truly understand large language models — from the math floor up to the frontier — as a single guided roadmap.
A layered map of what to learn to understand LLMs completely: the math & ML foundations, why transformers exist, the architecture, pretraining, alignment, inference & deployment, real-world use, evaluation, and the frontier. Some layers link to full interactive editions already in this library; others are explainer pages; the rest are on the way.
Learn top-to-bottom from 0. Available layers link to a built page or an interactive edition in this library; the rest are on the way.
The autoregressive objective at web scale: data, scaling laws, optimization, and distributed training across thousands of GPUs.
Turning a text predictor into an assistant: instruction tuning, RLHF, DPO, reasoning training, and LoRA fine-tuning.
In-context learning, prompting, RAG with embeddings and vector search, tool use, agents, and choosing prompt vs RAG vs fine-tune.
Benchmarks, perplexity, LLM-as-judge, and the failure modes: hallucination, bias, jailbreaks, and prompt injection.
Mixture of Experts, multimodal models, new architectures (Mamba/SSMs), and test-time compute / reasoning models.