Large Language Model (LLM)
Almost every LLM in use today is built on the transformer architecture from the 2017 paper Attention Is All You Need. The model reads text as tokens, and during training it learns to predict the next token across a huge body of text. The 2020 GPT-3 paper, Language Models are Few-Shot Learners, showed that a large enough model could pick up a new task from a few examples in the prompt without retraining, which is now called in-context learning.
After that first training stage, most assistants are tuned further to follow instructions and to decline harmful requests. Three limits shape how you use one. What it knows stops at its knowledge cutoff. It can only read as much as its context window holds. And it can state wrong things with confidence, which is called hallucination. Treat its output as a strong draft that you check, and give it the documents it needs through retrieval instead of relying on its memory.
Example
I'm going to paste a few rows from a sales spreadsheet.
First, tell me in three sentences what you can and can't reliably do with data like this.
Then ask me one question about what I want to learn from it.
{paste the rows here}When to use it
- Drafting, rewriting and summarizing text
- Sorting, tagging or extracting fields from messy documents
- Explaining a concept or a document in plainer words
- Any task where a person or a test can check the result
When not to use it
- Exact arithmetic over many numbers; give the model a calculator or code tool instead
- Questions about recent events with no retrieval or search connected
- Final decisions in high-stakes areas with no person reviewing them
Articles that use this
- Prompt Caching and Reuse: Design Prompts That Actually Hit the Cache at ScaleTechniques
- XML Tags Are the New Markdown: Why Closed Delimiters Beat Free-Form PromptsTechniques
- Stop Trusting Free-Form Chain-of-Thought. Make Your Model Sign a Logical Certificate InsteadTechniques
- Marc Andreessen Told His AI to 'Never Hallucinate.' Here's the Four-Line Scaffold That Actually WorksTechniques
- Chain-of-Thought Assumes Left-to-Right. Mercury 2 Doesn't: How to Prompt a Diffusion LLM by Handing It a Skeleton, Not a TrailTechniques

