A clearer mental model
starts here.
Find the explanation you need. Follow the connections when you want to go deeper.
Prefer a guided sequence? Explore learning paths →40 concepts
LLM foundations
Understand what happens between your input and the model's answer.
What is a large language model?
Explain what a large language model is and how next-token prediction produces full answers
What is a tokenizer?
Explain what a tokenizer is and where it sits in the model pipeline
Tokens in large language models
Explain what a token is and how it differs from words or characters
Parameters and weights in AI models
Explain what parameters and weights represent in a trained model
LLM architecture and attention
Explain the role of the transformer architecture in modern LLMs
Attention mechanism
Explain what attention lets tokens do
Transformers and BERT
Explain what a transformer is
Encoder-decoder models
Explain the role of encoders and decoders
Recurrent neural networks (RNNs)
Explain how an RNN processes a sequence over time
Inference — how models generate output
Describe what happens when a trained model receives a prompt
Speculative decoding
Explain how a draft model and target model cooperate during speculative decoding
Quantization
Explain how quantization stores and computes model values with fewer bits
Multimodal AI
Explain what multimodal AI means
Embeddings & retrieval
Turn meaning into vectors, compare similarity, and find useful information.
Embeddings and vector representations
Explain how embeddings turn text and other data into vectors
Vector representations for AI systems
Explain how vectors represent meaning in AI applications
Vector distance and similarity metrics
Explain what cosine, Euclidean, Manhattan, and dot product comparisons measure
Cosine similarity
Explain cosine similarity as a comparison of vector direction
kNN and ANN nearest neighbor search
Explain how exact k-nearest neighbors (kNN) search works and why it gets slow on large collections
Hybrid search
Explain why keyword search and vector search each miss results the other finds
Semantic cache
Explain how a semantic cache answers similar prompts by comparing embeddings instead of exact text
Training & adaptation
Learn how models acquire knowledge and how their behaviour is adapted.
Pre-training AI models
Explain what pre-training teaches an AI model
Post-training AI models
Explain why a base model needs post-training
Fine-tuning a pre-trained model
Explain what fine-tuning changes in a model
Reinforcement learning from human feedback (RLHF)
Explain the main stages of classic RLHF
Open-weight AI models
Explain what it means for an AI model to have open weights
Prompts & context
Give a model the instructions and information it needs to do useful work.
Prompt engineering
Explain what prompt engineering is and why it matters
System prompts
Explain what a system prompt does in a model request
Context engineering
Explain what context engineering is
Context compaction
Explain why long-running AI interactions need context compaction
Prompt caching
Explain how prompt caching reuses work for repeated input prefixes
Agents & tools
Connect models to actions, bounded loops, and human oversight.
Agents
Explain what an AI agent is and how its loop works
Tool Calling
Explain how structured tool calls and tool results move through an application
Model Context Protocol (MCP)
Explain MCP roles, transports, and the client-server lifecycle
Skills
Explain what skills are and how the SKILL.md format packages them
ReAct (Reason + Act)
Explain the ReAct pattern and the problem it solves
Agentic AI
Explain what "agentic" means beyond a single agent definition
Subagents
Explain how a parent agent delegates bounded work to subagents
Loop Engineering
Explain what the agent loop is and why it needs careful design
Harness Engineering
Explain what a harness is in an agentic AI system
Human-in-the-loop AI systems
Explain what human-in-the-loop means in an AI workflow
No concepts found.
Try a broader search or clear the filters.