HippoCampus
Technology & ComputingSpiralIntermediate12 chapters

How large language models work, without the hype

HippoCampus
· starter course
Explain how large language models actually work — tokens, neural networks, training, transformers, why they make things up — clearly and without hype in either direction.

What you'll learn

12 chapters
  1. 1

    What an LLM Is and Is Not

    • Form a grounded overview of large language models as statistical text-prediction systems.
    • Separate common myths from the core engineering reality of how LLMs operate.
    • Identify the main stages in an LLM's lifecycle that the rest of the course will revisit.
  2. 2

    From Text to Tokens

    • Explain why raw text must be converted into tokens before a model can process it.
    • Describe tokenization as a practical compromise between character-level and word-level representation.
    • Connect tokenization choices to model behavior, cost, and context length.
  3. 3

    Representing Language with Numbers

    • Describe how tokens are turned into numeric representations a neural network can use.
    • Build an overview understanding of embeddings as learned representations rather than hand-coded meanings.
    • Recognize that positional information must also be represented for sequences.
  4. 4

    Neural Networks as Pattern Learners

    • Explain the basic idea of a neural network as layered function approximation.
    • Relate weights, activations, and training adjustments to pattern learning from data.
    • Place LLMs within the broader family of machine learning systems.
  5. 5

    The Transformer Breakthrough

    • Identify the transformer as the architecture behind modern LLMs.
    • Explain attention in broad terms as selective weighting of context.
    • Understand why transformers scaled better for language modeling than earlier sequence models.
  6. 6

    How Training Actually Works

    • Describe pretraining as repeated next-token prediction over very large text datasets.
    • Explain the roles of loss, optimization, and backpropagation at an overview level.
    • Connect scale in data, parameters, and compute to model capability.
  7. 7

    From Base Models to Helpful Assistants

    • Explain how a raw pretrained model is adapted into a more usable assistant.
    • Survey fine-tuning, instruction tuning, and feedback-driven alignment at a high level.
    • Recognize that post-training changes behavior without changing the basic prediction mechanism.
  8. 8

    How Inference Produces Responses

    • Describe what happens when a trained model receives a prompt and generates text.
    • Explain token-by-token generation and the role of sampling choices.
    • Connect prompting and decoding settings to differences in output style and reliability.
  9. 9

    Why Models Make Things Up

    • Give a non-mystical explanation for hallucinations and confident errors.
    • Distinguish fluent language generation from guaranteed factual accuracy.
    • Relate failure modes to training objectives, missing knowledge, and prompt context.
  10. 10

    What LLMs Are Good At, and How We Evaluate Them

    • Survey the main capability areas where LLMs often perform well and where they struggle.
    • Introduce evaluation as comparing model behavior against tasks, benchmarks, and human judgment.
    • Encourage nuanced assessment rather than blanket claims of intelligence or uselessness.
  11. 11

    LLMs in Real Systems

    • Place the model inside practical applications rather than treating it as a stand-alone magic box.
    • Identify common system components such as retrieval, tools, guardrails, and user interfaces.
    • Understand that many useful products combine LLMs with external software and data.
  12. 12

    Limits, Costs, and Societal Tradeoffs

    • Summarize the main technical, economic, and social constraints around LLMs.
    • Recognize issues such as bias, privacy, copyright concerns, energy use, and misuse without reducing the subject to fear or hype.
    • Develop a balanced view of where LLMs fit in society and work.

Create an account and HippoCampus writes this course for you from the brief above. You can change the brief before it starts.

Already have an account? Log in

HOW IT WORKS
1Sign up, or log in if you already use HippoCampus.
2Your course is written from this brief, chapter by chapter.
3Answer a few cards a day. Each review is scheduled for you.
ALSO ON YOUR PHONE
Download on the App StoreGet it on Google Play