LLM Explainer Series

A plain-language series unpacking why modern models use Transformers, what attention does, how LLMs are trained, and how we should evaluate them.

Phase 1 · Transformer fundamentals

What problem did the Transformer actually solve?

From the sequential limits of RNNs to attention and parallel training.

An intuitive explanation of Query, Key, Value, and attention weights.

Multi-Head Attention, FFN, residual connections, and LayerNorm — what each component actually does.

Phase 2 · LLM training

Pretraining, instruction tuning, preference alignment, and what each stage is for.

Phase 3 · Model evaluation

A map of benchmarks, human preference, real tasks, and agent evaluation.