I keep confusing these

Each pair gives the one-sentence difference, the mechanical difference, an example and a trap question. These connections matter more than definitions.

34 pairs. For each: say the difference aloud first, then reveal.

Token vs Token ID

Token vs Embedding

Embedding vs Hidden state

Hidden state vs Latent representation

LayerNorm vs RMSNorm

Q vs K vs V

MHA / MQA vs GQA

Attention vs MLP

RoPE vs Absolute positional embedding

Logit vs Probability

Training vs Inference

Prefill vs Decode

Base model vs Instruct model

SFT vs RLHF

PPO vs DPO

RAG vs Fine-tuning

Embedding model vs Generative model

Vector DB vs Model memory

Tool vs Agent

Agent memory vs KV cache

Prompt injection vs Jailbreak

Prompt injection vs Data poisoning

Jailbreak vs Adversarial example

Model red teaming vs Penetration testing

Model poisoning vs RAG poisoning

Model extraction vs Training-data extraction

Authentication vs Authorization

Guardrail vs Sandbox

System prompt vs Security boundary

MCP client vs MCP server

Excessive agency vs Tool misuse

Data poisoning vs Context poisoning

KV cache vs Model memory (weights)

Agent memory vs Context window