AI concepts made simple
What a language model is, why it sometimes makes things up, what actually happens when you ask it a question. How AI works, explained in plain words — no formulas, no jargon — including the legal questions you run into along the way.
How AI works 34
- Should I give the AI examples or not? (zero-shot and few-shot)Zero-shot and few-shot: when to give the AI examples. For common tasks just ask; for precise format and style, two or three examples raise the quality a lot.
- What is "bias" in AI (and why does it concern me)?What AI bias is: a systematic distortion inherited from human data, presented in a neutral tone. How to spot it and ask for different perspectives.
- Can I use AI for health and wellbeing, and how far?Using AI for health and wellbeing: great for understanding, habits and preparing the visit. It doesn't diagnose or prescribe; for serious signals, the doctor.
- How does the AI generate "creative" ideas (and is it really creativity)?How the AI generates creative ideas: it recombines what it has seen into unexpected combinations, in bursts. The choice of value stays yours.
- How does AI remember what we said in the chat (and why does it sometimes forget)?How memory works in AI chats: it rereads everything at each turn, but the context window is limited and the beginning drops out. Here's why it forgets.
- What is the AI's step-by-step reasoning (chain of thought)?What step-by-step reasoning is (chain of thought): making it write the steps before the answer reduces errors on logic and calculations.
- How does the AI behave when I ask for personal advice or touch on sensitive topics?How the AI handles personal advice and sensitive topics: it helps you reason but doesn't decide. For serious distress it refers you to a real person.
- What is AI "alignment" (and why is everyone talking about it)?What AI alignment is: making AIs behave helpfully, honestly and harmlessly. It explains refusals, tone and limits. It reduces risks, doesn't zero them.
- What is the system prompt, the instructions the AI receives before me?What the system prompt is: the hidden sheet of instructions the AI receives before you, governing tone and limits of every answer. It's not code, it's language.
- Why does the AI repeat the same things or wander off topic?Why the AI repeats or wanders off topic: a chat that's too long overloads the context and the key points stay buried. The cure: start over from a clean chat.
- What is "self-consistency" and why can asking several times give better answers?What self-consistency is: generating several independent answers and keeping the one that recurs. On problems with a correct solution it reduces random errors.
- How do the usage limits of free AIs work (and how to live with them)?How free AI limits work: a budget of messages on the best model, then a lighter version or waiting. How to live with them well.
- What is "prompt engineering" in simple terms?What prompt engineering is in simple terms: writing requests well for better answers. It's not code, it's clarity: role, context, examples.
- What is the AI's "temperature" and should I change it?What the AI's temperature is: it governs how much it dares with words, low is precise, high is creative. In the apps it's fixed: you get the effect with words.
- Why do some AIs follow complex instructions better and others get lost?Why some AIs follow complex instructions better: it depends on training and how much you ask at once. Number, don't contradict yourself, break it up.
- What is "prompt injection" and should I worry about it?What prompt injection is: instructions hidden in a text that hijack the AI. The risk grows with assistants that act. How to defend yourself.
- How does ChatGPT (or Gemini) remember me from one conversation to the next?AI long-term memory: a notebook about you, separate from the chats, reread at every conversation. How to see it, correct it and delete it.
- What are custom instructions and how do I set them up well?What custom instructions are: fixed rules you set once and the AI applies to every chat. Two boxes, a few clear and compatible rules.
- What is prompt chaining (linking requests together)?What prompt chaining is: breaking a task into steps and using the result of one as input for the next. Each step comes out better and you check halfway through.
- What does it mean that an AI is "multimodal" (and what do I do with it)?What multimodal AI is: it understands text, images, audio and sometimes video, and connects them. Photograph instead of describing, talk instead of writing.
- Why do some AIs answer better in Italian (and others less)?Why some AIs answer better in Italian: they learn mostly from English. On hard tasks English remains the best ground.
- Why does the AI give me generic answers, and how do I get more detail?Why the AI gives generic answers and how to get more detail: give it context, constraints and an example. The generic is the mirror of a vague question.
- Why does the AI sometimes refuse to do what I ask?Why the AI refuses certain requests: filters and guardrails block the harmful, sometimes the lawful too. For false positives, just clarify context and purpose.
- Why won't the AI tell me where to invest, and what is it good for then with money?Why the AI gives no tailored financial advice: it knows neither you nor the future of the markets. It can explain the instruments, do the math and your budget.
- Why does the AI sometimes ask me for clarification, and how do I ask better questions?Why the AI asks for clarification and how to ask better: give purpose, recipient, format and constraints. The vague request is the cause of errors.
- How do I tell whether I can trust an AI's answer?Checking whether an AI's answer is reliable: open the sources, check numbers and names, turn on web search. The final judgment stays yours.
- What happens to my AI chats? How do they handle data privacy?How AIs handle chat privacy: by default they can train on your data, even paid. Turn off the option and don't write secrets.
- Why do some AI answers seem so natural and "human"?Why AI answers seem human: it's trained on our language and tuned to sound natural. It imitates the form, it doesn't understand or feel anything.
- What is "context switching" when I talk to the AI (and why does it confuse me)?What context switching is: changing topic in the same chat confuses the AI, which rereads everything. The cure is simple: new topic, new chat.
- Why do AIs sometimes make up information (hallucinations)?Why AI invents information: it doesn't verify, it predicts the most probable word and fills gaps with plausible but false details. What hallucinations are.
- Why do some AIs stay more consistent and others get lost along the way?Why some AIs stay more consistent: training, context width and how full the chat is. But consistency is also how you manage the conversation.
- Why does the AI give me one answer and shortly after the opposite one?Why the AI contradicts itself: it has no stable convictions, generates each answer on the spot, is sensitive to how you ask and tends to go along with you.
- Why doesn't the AI give me precise medical or legal advice, and what can it do then?Why the AI doesn't give tailored medical or legal advice: an educational role, not a consultant. It explains concepts and prepares your questions for the pro.
- What Is an LLM, Explained Simply
Practical legal matters 13
- Comparing transcription tools: which one is most accurateWhich transcription tool is most accurate depends on the audio more than the software: what accuracy really measures and how to improve the result.
- Can AI-generated images be used freely?Using an AI-generated image and owning it are different things: what the tools' terms and copyright say, and how not to take risks.
- Comparing tools to enhance and restore photosTools to enhance and restore photos compared: enlarge or repair, web or software. The common limit and why the starting scan matters.
- What are the legal risks of AI with sensitive dataThe legal risks of AI with sensitive data: retention, reuse for training, exposure. What not to input and how to protect third-party data.
- How to use AI-generated content at work without risksUsing AI content at work without risks: close the three dangers (confidential data, invented facts, copies) with a fixed verification process.
- Can you publish a text written by AI?Publishing a text written by AI is legal, but with three cautions: the declaration platforms require, ownership and your responsibility.
- Who is responsible if AI gives wrong information?If AI gives wrong information the responsibility is whoever uses or offers it, not the AI's. Who really answers and how to verify before acting.
- Who owns a text or an image created with AI?Who owns a text or image created with AI: a pure output often belongs to no one, it becomes yours as much as the creative work you add to it.
- Which automatic subtitle tool is more reliableThe most reliable subtitle tool isn't just the most accurate: timing, languages, and ease of correction matter. How to choose based on where they end up.
- Free or paid image generators: which to chooseFree or paid image generators: free is enough to learn, paid is for volume, consistent quality and a commercial license.
- Can AI infringe the copyright of existing works?AI can infringe copyright in two ways: how it was trained and what it returns to you. When the risk becomes yours and how to check before publishing.
- How to cite AI in a document, a thesis, or an articleHow to cite AI in theses, documents, and articles: APA, MLA, and Chicago formats, what to indicate, and why you must first check your school's rules.
- Searching with AI or with Google: when one or the other is betterSearching with AI or Google: AI gives a chewed-over answer, Google gives the sources. When to understand fast and when you need to verify yourself.