Reading about large language models only goes so far. A good video can show the shape of an idea, such as how attention lets each word look at others, in a few minutes. This list collects six free videos from two of the most respected explainers on the subject, 3Blue1Brown and Andrej Karpathy, and puts them in an order that builds understanding step by step. We checked that each link points to the video named; we describe what the videos cover at the level of their titles and well-known contents, and we do not claim they are the only good options. For more picks, see our lists of essential AI talks and lectures and YouTube channels to learn AI.

1. Large Language Models explained briefly (3Blue1Brown)

Best for: absolute beginners who want a short, visual overview first.

This short video from 3Blue1Brown, the math-animation channel run by Grant Sanderson, introduces what a language model is and how it predicts text, and it works as a primer for the longer chapters. Watch it first to get the vocabulary: model, training, prediction. It is the right place to start if terms like token or transformer are new.

2. But what is a neural network? (3Blue1Brown, Deep Learning chapter 1)

Best for: anyone who wants to understand the building block underneath every modern model.

The first chapter of 3Blue1Brown’s deep learning series explains neural networks using the classic example of recognizing handwritten digits. It shows layers of simple units, how weights and biases shape what the network detects and why the structure can learn patterns. It gives you the picture behind our glossary entry on neural networks. Chapter 2 of the same series, on gradient descent, explains how such networks learn, and is a natural follow-up if you want the training side.

3. Transformers, the tech behind LLMs (3Blue1Brown, Deep Learning chapter 5)

Best for: viewers ready to see how a language model is actually built.

This chapter walks through the transformer architecture at the level of data flowing through the model: text becomes tokens, tokens become vectors and layers of processing turn those vectors into a prediction for the next token. It is a conceptual tour, not a coding lesson. Watch it before the attention video, because it sets up the pieces that attention then explains.

4. Attention in transformers, step-by-step (3Blue1Brown, Deep Learning chapter 6)

Best for: people who keep hearing “attention is all you need” and want to know what it means.

Chapter 6 focuses on the attention mechanism, the idea that lets each position in a sequence weigh information from others. If you read our glossary entry on the attention mechanism and found it abstract, this video’s step-by-step animation is the best complement. Expect to pause and rewatch; it is dense, and that is normal.

5. Intro to Large Language Models (Andrej Karpathy)

Best for: a one-hour, big-picture talk that covers how models are trained and used, including safety topics.

Karpathy, a well-known AI researcher and former director of AI at Tesla and a founding member of OpenAI, gives a general-audience talk on what large language models are, how they are trained and how they are used. The talk also discusses security issues such as jailbreaks and prompt injection, which connect directly to our entries on jailbreaks and prompt injection. It is a good bridge between visual intuition and the practical picture of how chatbots come about.

6. Deep Dive into LLMs like ChatGPT (Andrej Karpathy)

Best for: viewers who want the full pipeline from raw data to a chatbot and have a few hours.

This long video goes through the stages used to build a chat assistant: gathering and processing text, pretraining, then post-training that shapes behavior, with discussion of why models hallucinate and how reinforcement learning fits in. It is aimed at a general technical audience and rewards patient viewing in several sittings. If you want to see the code side after this, Karpathy’s “Let’s build GPT: from scratch, in code, spelled out” builds a small transformer step by step, and suits viewers who already program in Python.

A suggested watching plan

  1. Day one: videos 1 and 2.
  2. Day two: videos 3 and 4, taking notes on tokens, vectors and attention.
  3. Day three: video 5 for the big picture and risks.
  4. When ready: video 6, over a weekend, then try small experiments yourself.

How to get the most from them

Pause often and explain each idea aloud in your own words. Re-watch the attention chapter after the Karpathy talks; it makes more sense in context. Treat any video as a snapshot: models and training methods change quickly, so check recent sources for current details.

What to watch next

  • Newer chapters or updates to the 3Blue1Brown series on language models.
  • More long-form lectures from researchers on post-training and reasoning models.
  • University course lectures that go deeper into the math, such as those in our Math for AI pieces.
  • Hands-on follow-ups where you train or run a small model yourself.

Key takeaways

  • Start with short visual explainers before long technical talks.
  • 3Blue1Brown excels at intuition for neural networks, transformers and attention.
  • Karpathy’s talks cover how chatbots are trained and the risks around them.
  • Watch actively, pause and explain each idea in your own words.

Frequently asked questions

Do I need math to follow these videos?

Not for the first three or four; they are built for visual understanding. The longer Karpathy video assumes some technical patience, not advanced math.

Are these videos up to date?

The fundamentals, such as neural networks, transformers and attention, change slowly. Details of the newest training methods move faster, so supplement with recent articles.

Where should I go after these?

Try running a small open model locally or work through a hands-on tutorial. Our guide to running a local model with Ollama is a good first step.