- How Many Tokens Is a Word? Tokens Explained with Real Counts
A token is a chunk of text, often a word or part of one. See real token counts for English, other languages and code, and why counts differ between models. - How Much VRAM Do You Need to Run an LLM Locally? (With the Math)
VRAM needed = parameters × bits per weight ÷ 8, plus KV cache and runtime overhead. A table for 8B to 70B models at FP16, 8-bit and 4-bit, with worked examples. - How to Reduce AI Token Usage: 8 Levers and Their Trade-offs
Cut LLM token costs by trimming context, summarising chat history, prompt caching, smaller models, output limits and batching. How each works, and what it costs you. - How to Run an LLM Locally with Ollama (Windows, macOS, Linux)
Install Ollama, then run a model with one command: ollama run gemma4. Install steps per OS, pull, list and ps commands, picking a model size, and the local API. - How to Write a Good AI Prompt (With Before and After Examples)
A good prompt states the task, context, constraints and output format, and shows an example. Four before/after prompts based on OpenAI and Anthropic guides. - System Prompt vs User Prompt: Roles in Chat APIs Explained
The system (or developer) prompt sets the app's standing rules; user messages carry each request. What goes where, with request JSON for OpenAI and Claude. - Temperature and Top-p Explained: What LLM Sampling Settings Do
Temperature reshapes next-token probabilities; top-p cuts off unlikely tokens. Worked numbers, each API's ranges, and why newer Claude and Gemini models drop them. - What Is a Context Window? Limits, Overflow and Long-Context Costs
A context window is the maximum number of tokens a model can handle in one request, input and output together. What happens when you exceed it, and why it matters. - What Is LLM Quantization? GGUF, Q4_K_M and Q8_0 Explained
Quantization stores model weights in fewer bits so an LLM fits in less memory. What GGUF is, what Q8_0, Q6_K and Q4_K_M mean, and how much quality each costs. - Why AI Hallucinates, and 7 Ways to Reduce It
AI models give confident wrong answers because training and testing reward guessing. Practical fixes: give sources, allow "I don't know", require quotes, verify.
AI & prompting
How language models work, writing prompts, tokens, context windows and running models locally.