Reference
How this actually works
Written for someone who wants to understand the thing, not be sold it. Same articles in Arabic and in English, written twice rather than translated once.
What we have had to understand to build in two languages, written down properly. No product pitch — these are here to be useful whether or not you ever use anything of ours.
- What Makes Arabic Hard for Language ModelsTokenisation, root-and-pattern morphology, missing diacritics, dialects and thin data: the real reasons Arabic costs a model more than English does.
- Writing a Prompt That Works — Including in ArabicThe practical craft of prompting, plus the Arabic-specific parts: choosing your prompt language, pinning dialect and register, and what length costs.
- Why a model makes things up, and what actually reduces itA model produces likely text, not checked facts. Where it invents most, why a confident tone proves nothing, and which fixes actually reduce the errors.
- What a context window is, and why long chats driftWhy a model re-reads the whole conversation each turn, what happens when the window fills, and how to stop your long chats from drifting off course.
- What an AI Agent Actually Is, and What It Is NotThe real difference between a model that answers and a system that acts: the loop, tools, verification, compounding failure, and what a human must keep.
- Getting Trustworthy Answers Out of Your Own DocumentsWhy a general model invents answers about your PDF, how retrieval narrows it, where chunking and OCR fail, and what Arabic text adds to the problem.
- What AI Is Actually Good At in Code, and What It CostsA concrete look at the coding tasks language models handle well, the ones that quietly cost you hours, and the review habits that keep the two apart.
- When a bigger model earns its cost, and when it doesn'tBigger model tiers buy depth on hard problems and nothing on easy ones. What actually predicts needing one, what does not, and how to test it yourself.
- How to Read an AI Benchmark Score, and When to Distrust ItWhat benchmark scores measure, why contamination and self-reporting distort them, and how to tell whether a number predicts anything for your work.
- What happens to what you type into an AI toolWhat providers do with your prompts, uploads and screenshots: training defaults, retention, human review, team workspaces, and where data is processed.