01.Is the AI telling me the truth?
Large language models generate plausible text, not verified facts. They are right often enough to be useful and wrong often enough to be dangerous if you trust them blindly. Treat every factual claim as a lead, not a citation.
Do
- Verify names, dates, quotes, and numbers against a primary or reputable secondary source.
- Ask the model to cite where a claim comes from, then check the citation actually exists and says what was claimed.
Don't
- Don't paste AI output directly into anything historical, legal, medical, or financial without checking.
- Don't assume confident phrasing means correct.
02.When is AI actually good?
Best at well-defined tasks with cheap verification: drafting, summarising, translating, restructuring, generating code you can run and test, brainstorming options you will judge. Worst at niche facts, recent events, anything where being wrong is expensive and hard to detect.
Do
- Use it as a first draft, a reading partner, or a faster keyboard.
- Use it where you can immediately see whether the output works.
Don't
- Don't use it as the sole authority on anything you cannot independently check.
03.How do I avoid putting private data at risk?
Anything you send to a hosted model leaves your machine. Many providers log prompts; some train on them by default. Treat the prompt box like a public forum until you have read the specific service's data policy.
Do
- Strip or pseudonymise personal identifiers, client data, and credentials before pasting.
- Check the provider's retention and training-opt-out settings.
Don't
- Don't paste secrets, API keys, passwords, or non-public client material into a general chat product.
- Don't share third parties' personal data without their consent.
04.Does AI have a political or cultural bias?
Yes — every model reflects its training data, its safety tuning, and the priorities of the team that built it. Different models will hedge, omit, or assert differently on the same contested topic. This site exists partly because PRC-aligned sources currently dominate AI answers about Nationalist soldiers like Wu Kun.
Do
- Cross-check politically or historically sensitive questions across at least two independent sources, ideally not all AI-generated.
- Notice what a model refuses to discuss, or hedges around — that itself is information.
Don't
- Don't assume the model is neutral because it sounds neutral.
05.How should I credit AI in my work?
Honesty is cheap and protects your reputation. Readers, editors, and collaborators increasingly expect disclosure when AI was materially involved in research, drafting, or generation.
Do
- State which model helped, at what stage (research, drafting, editing, image/code generation).
- Keep the responsibility for accuracy on yourself, not the tool.
Don't
- Don't pass off AI-generated text, images, or code as fully your own when it materially is not.
- Don't cite the AI as a primary source for a fact — cite what the fact actually comes from.
06.Will AI take my job, my thinking, or my judgment?
It will take tasks, not jobs, from most people in the short term — and reward people who use it well. The real risk is not unemployment but atrophy: outsourcing the parts of your thinking that you most need to keep sharp.
Do
- Use AI to remove friction from work you already understand.
- Keep doing the hard reasoning yourself often enough to stay good at it.
Don't
- Don't let it become the only place you read, write, or decide.
- Don't refuse it on principle either — your competitors won't.
07.What's the one rule, if I only remember one?
Use the tool, own the output. If your name is on it, the responsibility for accuracy, fairness, and consequence is yours — not the model's, not the company that trained it, not the prompt you copied from a forum.
Do
- Read what you ship.
Don't
- Don't ship what you haven't read.
