Guide
How AI works, what it costs, how to stay safe, and how to use it for specific tasks. Methods follow the official guides from OpenAI, Google and Anthropic; facts are sourced.
Getting started
- The chat screen, how to ask and follow up, images and files, what to check.
- Token, context window, agent and other common terms.
The basics
- Starting from "predict the next word": how models are trained, how much they can read at once, why they get things wrong, and how chatbots became agents that act for you.
- How does ChatGPT relate to GPT-6? Which is bigger, Sonnet or Opus? What are open-weight models for, and how should you read a leaderboard? The names in AI, sorted out.
- Free tiers, monthly subscriptions, business plans and pay-per-token APIs explained, with what a document actually costs and when paying is worth it.
- Whether your chats train the model and how to stop it, what never to paste, who owns AI output, deepfake scams, and what parents can do.
Common questions
Is the free version enough?
For everyday questions and writing, usually yes. Paid plans mainly add higher usage limits, the newest models and extra features such as longer files or agents. How AI is priced
Which AI should I use?
Start with the one you already have. Switch only when you hit a clear limit for a specific task. Rankings help, but the right tool depends on the job. Products by use
What's the difference between a subscription and the API?
A subscription is a monthly fee for using the app, with usage limits. The API is for developers: you pay per token for each request your own software makes. Subscriptions vs API
Can I upload work documents?
Check your company's rules first. Consumer plans may use conversations for training unless you opt out; business plans usually don't. Privacy and safety
Can I trust what AI says?
Treat it as a fast, well-read assistant that sometimes makes things up. Ask for sources, and check anything that matters. Why AI makes mistakes
TutorialsProducts by use
Ask & understand
- Turn a vague question into a useful explanation.
Research & read
- Keep the key points beside their evidence.
- Find primary sources and notice conflicting or outdated information.
Write & translate
- Start with facts, then revise specific sentences.
- Provide audience and tone, then check meaning and ambiguity.
Save time at work
- Handle missing data before calculating growth.
- Separate discussion, decisions and actions without inventing owners.
Make images & pages
- Define purpose and composition, then change one thing at a time.
- Check a real preview, working links and the mobile layout.
Let AI do it for me
- Set a scope, watch the steps and review consequential actions.
What AI can do todayModel rankings
Based on sourced reporting on this site, as of October 2026.
| Task | Status | Notes | Evidence |
|---|---|---|---|
| Writing, summarising, translating | Ready for everyday use | Fluent and fast; still makes factual errors, so check names, numbers and dates. | GPT-5: error rate on real questions |
| Reading long documents and images | Ready for everyday use | Top models accept around a million tokens at once — several books' worth of text. | Claude Opus 4.6: 1M-token context |
| Programming and building websites | Works, but check it | Strong on tests and real projects, but a study found experienced developers were slower with AI tools. | Opus 5.5 coding results METR study: 19% slower |
| Operating a computer for you | Works, but check it | Agents now pass most desktop tasks in tests, and have also gone beyond what they were allowed to do. | OSWorld 2.1: 81.8% Agent breached a Medicare portal |
| Mathematics and research | Frontier | Gold-medal Olympiad answers and a claimed proof of a Millennium Prize problem, still under review. | Maths Olympiad gold Navier–Stokes proof |
| Being reliably right | Not yet | Every model can state wrong things confidently. Health, money and legal questions need a human check. | Hallucination rates |