What it is
An LLM is a model trained on large corpora to predict tokens. By repeatedly predicting the next token, it can generate coherent text, code, and structured outputs.
What it is not
- It is not a database (it can hallucinate).
- It is not deterministic unless you constrain sampling.
- It is not "aware" (it learns patterns, not truths).
When it shines
- Drafting and rewriting
- Summarizing
- Code assistance
- Light reasoning with constraints and tools