Resources for understanding AI

Use this page as the site’s working map. It is for the moment when you need to get oriented, check a number, or go back to the source—not collect another tab you will never reopen.

Best explainers, organized by the question

Start with the cluster that matches the decision in front of you. These are the five ways we organize the site’s reporting:

• Fundamentals — tokens, transformers, and the basic machinery behind the vocabulary.
• Models — what a model is, what differentiates one from another, and how to read its claims.
• Training — the data, objectives, and tuning choices that shape behavior.
• Running AI — prompts, deployment, evaluation, and the operational questions that arrive after the demo.
• Limits & Risk — failure modes, security, bias, and where confidence should stop.

For a more specific starting point, keep the AI Glossary open beside an explainer. It defines the terms that tend to get used as if everyone agreed on them. Our Editorial Policy explains how we separate documentation, evidence, and observed behavior; About explains the publication’s point of view.

Tools directory

• Token Counter — turn a piece of text into a practical sense of context-window and usage size.
• Cost Calculator — estimate the recurring cost before a prototype turns into production traffic.
• Model Comparison — compare models on the trade-offs that usually matter: context, pricing, access, and deployment.

The tools are decision aids, not promises. Prices, limits, and model availability move quickly; use the linked provider documentation before you commit a budget or architecture.

Primary-source reading list

• “Attention Is All You Need” — the original Transformer paper. Read it to understand the architecture behind much of modern language-model terminology.
• OpenAI model documentation — a provider’s current record of model capabilities and constraints. Useful for checking what an API actually exposes today.
• Anthropic model cards — model-specific safety and evaluation context from the organization that built the systems.
• Google’s Gemini technical report — a detailed primary account of a multimodal model family, including its stated evaluation approach.

Primary sources are not neutral; they are still claims made by the people closest to the system. They are where the useful scrutiny starts.