What is Benchmark?
Benchmark: a fixed test suite scoring models on the same tasks so results compare. What good suites measure and where they mislead.
1 min read↗
Plain words first
Glossary.
Short definitions of the ideas this journal keeps returning to. 15 of 34 terms published; a new one lands every weekday.
What is Context window?
A context window is the stretch of tokens a model can read at once. What sets its size, what happens past the edge, and how builders fit work inside it.
1 min read↗
What is Embeddings?
Embeddings turn text, images, or rows into number lists placing similar meanings together. How vectors are made and used for search.
1 min read↗
What is Fine-tuning?
Fine-tuning keeps training a ready model on focused examples for one task or style. What changes, what it costs, prompting vs tuning.
1 min read↗
What is Hallucination?
Hallucination: a fluent model answer that sounds true but is not grounded in sources. Why models invent facts and how builders stop it.
1 min read↗
What is Inference?
Inference turns a trained model plus a fresh prompt into an answer. How serving differs from training, latency and cost budgets.
1 min read↗
What is KV Cache?
KV cache reuses past attention keys and values so long chats and agents run faster. What it costs, and how builders shrink it.
2 min read↗
What is Prefix caching?
Prefix caching reuses stored attention work across requests that share the same opening tokens. How it hits, and how builders keep it hot.
2 min read↗
What is Prompt caching?
Prompt caching is a provider-billed discount for reusing a stable prompt prefix. What it caches, what it costs, and where the breakpoints go.
1 min read↗
What is prompt injection?
Prompt injection hides hostile instructions in data a model reads, hijacking its task. How the trick works and builder defenses.
1 min read↗
What is Quantization?
Quantization stores model weights in fewer bits to fit small GPUs and laptops. Accuracy-vs-speed tradeoffs and when builders use it.
1 min read↗
What is RAG?
RAG grounds a model in documents retrieved at query time instead of memory alone. How the retrieve-then-write loop works for builders.
1 min read↗
What is System prompt?
A system prompt is the hidden instruction block that sets a model's role, rules, and tone. What it controls and how builders shape it.
1 min read↗
What is Temperature?
Temperature is the sampling dial for token randomness. What low and high settings do, and where builders set it for fact vs flair.
1 min read↗
What is Vector database?
A vector database stores embedding vectors so search finds meaning, not keywords. How similarity indexes work and how builders use them.
1 min read↗
No definitions found.
Try a different word, or return to the full glossary.