Introduction to Artificial Intelligence: A Handbook for the Curious

A guide for everyday life, work and the digital future

The all-important question: search or ask?

Before we start, you're probably asking yourself a question:

"Should I type my question into Google or ask an AI like ChatGPT?"

The answer is: it depends on what you're looking for.

• Use Google (or other search engines) if:

  • You're looking for an official website (e.g. "Deutsche Bahn login").

  • you need the latest sports results or weather data in real time.

  • you want to buy a product and compare prices.

• Use an AI if:

  • You need a summary ("Explain the property tax reform to me in three sentences").

  • you need help with structuring ("Create an agenda for a golden wedding anniversary").

  • you are looking for a sparring partner for ideas ("Which gift would suit a hobby gardener?").

Remember: Google provides you with the sources (links), the AI provides you with the answer.

Important findings

• AI is accessible and useful: Models like Google Gemini, Microsoft Copilot, Claude, Grok and Perplexity offer versatile help for everyday tasks, but they are tools that should be used carefully to avoid misinformation.

• Understand basic terms: LLM (Large Language Model) are the core technology behind many AI chats; RAG (Retrieval-Augmented Generation) helps make answers more accurate by incorporating external data; Agents are smart helpers that automate tasks.

• Possible applications: AI is good for research, creative writing or problem solving, but for fact-based questions often better than search engines like Google as they explain context - it seems likely that a mix of both is optimal.

• Payment models (as of January 2025): Most premium plans cost around $20 per month, with free basic versions; there are variations depending on the provider, such as Grok over X Premium+ for around $22.

• Caution is advised: Hallucinations (false info) can occur; privacy is critical - don't share sensitive data; good prompts personalize usage and reduce risk, though the same prompts may yield different results for different users due to context and model variations.

• Controversy around AI: some see AI as a helpful tool, others warn of dependency or bias; evidence suggests responsible use brings benefits without ignoring all sides.

Overview of AI models and agents

AI models such as ChatGPT (from OpenAI), Google Gemini, Microsoft Copilot, Anthropic's Claude, xAI's Grok and Perplexity are advanced systems based on large language models. They process text input (prompts) and generate responses. AI agents are enhanced versions that not only respond, but can also perform tasks such as planning a trip or analyzing data autonomously. They are best suited for creative brainstorming, learning assistance or efficient research, where they explain complex relationships.

Payment models and prices (as of January 2025)

Many AI services offer free versions with restrictions, such as limited use or slower responses. Premium plans unlock faster models, more features and higher limits:

• Google Gemini: Basic free; Advanced/Pro approx. 19.99-20 US dollars/month (incl. integration in Google apps).

• Microsoft Copilot: Basic free of charge; Pro approx. 20 US dollars/month, Enterprise from 30 US dollars/user.

• Claude: Basic free of charge; Pro approx. 20 US dollars/month, Max from 100 US dollars for intensive users.

• Grok: About X Premium+ approx. 22 US dollars/month (potentially higher later); API usage cheaper.

• Perplexity: Basic free of charge; Pro approx. 20 US dollars/month with unlimited searches.

• ChatGPT: Basic free of charge; Plus approx. 20 US dollars/month, Pro from 200 US dollars.

These prices may vary by region and often include annual discounts.

Pitfalls and tips for private users

As a private user, you should pay attention to data protection: Avoid sharing personal data, as entries are potentially saved. Absolute no-gos are using AI for illegal activities or blindly trusting medical/financial advice. Check sources by asking AI to provide references and compare with reliable websites. Good prompts are clear, detailed and personalized - e.g. "Explain to me as a 40-year-old amateur cook how to optimize a simple recipe." Hallucinations are caused by gaps in training data; recognize them by inconsistencies and avoid them with specific queries.

When to use Google vs. AI?

Should I ask my question in Google or in an AI like ChatGPT? It depends: For actual facts or sources, Google is often more reliable as AI can hallucinate. Use AI if you need explanations, summaries or creative ideas - e.g. "How do I plan a trip?" vs. "Latest news on a topic".

Detailed exploration of artificial intelligence

This section dives deeper into the world of AI, explaining terms, concepts and practical applications in detail. It builds on the points above and provides examples, illustrations through tables and rhetorical questions to deepen understanding. Remember: AI is a tool that can enrich your life, but it requires conscious use.

Important abbreviations and technical terms in the AI environment

AI is full of jargon, which can seem overwhelming at first. Here is a comprehensive list with explanations, sorted alphabetically. Each term is described simply, with examples and why it is important.

• AGI (Artificial General Intelligence): A hypothetical AI that surpasses human intelligence in all areas. Current AI is "narrow" (specialized), AGI would be general. Important because it stimulates debates about the future of AI.

• AI Agent: An autonomous program that combines LLMs with tools to complete tasks. Example: An agent plans your shopping list, checks prices online and suggests recipes. Agents are useful for automation, such as managing emails or creating reports.

• Chain-of-Thought (CoT): A prompting technique in which AI thinks step by step. Example: "Solve the puzzle step by step." Helps to solve complex problems and reduces errors.

• Deep learning: An ML method with neural networks that process layers of data. Basis for modern AI, e.g. image recognition.

• Fine-tuning: Adapting an LLM to specific data. Example: Training a model for medical texts. Makes AI more specialized, but more expensive.

• Generative AI (GenAI): AI that creates content, such as texts or images. Examples: DALL-E for images, ChatGPT for stories.

• Grounding: Linking AI answers with real data to avoid hallucinations. Important for reliability.

• Hallucination: When AI invents false but plausible-sounding information. Reasons: Gaps in training data or overgeneralization. Recognize it by contradictions; avoid by RAG or source checking.

• Inference: The process by which a trained model processes new data and generates answers. Fast and inexpensive compared to training.

• Latency: The delay until an answer is given. Important for real-time applications such as chats.

• LLM (Large Language Model): A model trained on huge amounts of text to understand and generate language. Examples: GPT series. Core of chats such as Claude or Gemini.

• ML (Machine Learning): AI that learns from data without being explicitly programmed. Subset of AI.

• NLP (Natural Language Processing): Processing of human language by AI, e.g. translation or sentiment analysis.

• Prompt: The input you give the AI. A good prompt is clear, detailed and structured. Example: "Write an email as CEO to employees about team building."

• RAG (Retrieval Augmented Generation): Combines retrieval (searching for relevant data) with generation. Reduces hallucinations by incorporating external sources. Example: AI retrieves facts from a database before answering.

• Token: The smallest unit that AI processes (e.g. parts of words). Prices are often based on this; a prompt of 1000 tokens costs more.

• Transformer: Architecture behind LLMs that directs attention to important parts. Revolutionized AI since 2017.

Technical terms explained clearly

• LLM (Large Language Model): The statistical model behind AI. It has read billions of sentences and learned which word follows the next and with what probability.

• Prompt: Your instruction. The more precise the prompt, the better the result.

• RAG (Retrieval Augmented Generation): A process in which the AI first reads a reliable document (e.g. a company brochure) before responding. This minimizes errors.

• Chat window: The area in which you write. Tip: Think of it like a WhatsApp chat with a very smart intern.

• Hallucination: The AI invents facts. It's not deliberately lying, it's just calculating a word that sounds logical but is factually incorrect.

Table: Comparison of common technical terms

AI agents: What they are and what they're for

AI agents go beyond simple chats. They are like digital assistants that act: they call up tools, learn from interactions and complete sequences. Example: an agent in Copilot organizes your notes, searches for info and creates a report.

Best used for:

• Productivity: Automate routine tasks like email sorting.

• Learning: Explain concepts interactively, e.g. "Simulate a conversation with Einstein."

• Creativity: Generate ideas for hobbies, such as recipes or stories

Table: Areas of application for AI agents

Payment models in detail (as of January 2025)

AI services are financed by subscriptions, API usage or advertising. Free tiers are limited (e.g. slower models), premium offers priority.

• Subscription models: Monthly, e.g. 20 US dollars for unlimited use.

• Pay-per-use: Based on tokens, e.g. Grok API: 0.20 US dollars per million input tokens.

• Enterprise: For companies, from 30 US dollars/user, with data protection.

Table: Prices of the models mentioned

Prices may increase; check current offers.

Pitfalls when using AI: what to look out for?

AI is powerful, but not perfect. Here are detailed warnings for private users.

• Absolute no-gos: Do not share sensitive data (e.g. passwords, finances). Do not use AI for harmful purposes such as fraud.

• Check sources: Always ask: "Cite sources." Compare with Google or official sites.

• Write good prompts: Be specific, give context, assign roles (e.g. "As an expert, explain..."). Personalization: the same prompt ("Explain AI") gives a detailed answer for you (40+, interested), a simple one for a beginner - because models take into account context from previous interactions or make implicit assumptions.

• Hallucinations: Reasons: Incomplete data. Recognize: Inconsistent facts. Avoid: Use RAG-enabled tools, ask for evidence. Example: AI says "The moon is made of cheese" - check with facts.

Data protection: What you should never do

Data protection for private users

Data protection is crucial as AI input could be stored. Example: If you describe medical symptoms, anonymize them. Use VPNs or private modes. Rhetorical question: Should you enter your address into an AI or should you use a search engine? - The latter to avoid tracking. Best practices: Delete chats regularly, choose privacy-friendly providers like Claude.

The "no-go" list for private users

AI learns from what we feed it (unless you explicitly turn it off).

• Example 1 (work): Never copy an employment contract or secret strategy papers into the AI.

  • Better: "Here is a fictitious contract text. Can you explain the clause on the notice period in general terms?"

• Example 2 (private): Do not enter health data with clear names.

  • Bad: "My husband Max Mustermann has the following blood values..."

  • Better: "How should these blood values (anonymously) be assessed for a 50-year-old man?"

• Example 3 (finances): Don't post credit card numbers or PINs, even to take "tests".

Golden rule: Only enter information that you would also write on a postcard.

Why does AI respond differently to each person?

The secret of personalization

You may have noticed that when two people enter the exact same prompt, they often get different answers. Why is that?

1. The context (the memory): Most AIs "remember" the history of the current conversation. If I was talking about gardening before, the answer to the question "What do I need for spring?" will be different from someone who was chatting about fashion before.

2. Custom Instructions: In the settings, you can tell the AI, "I'm a teacher, always answer me educationally." This influences every answer in the background.

3. The temperature of the AI: AIs work with probabilities. They are programmed to add a certain amount of "creativity" (chance) so that they don't sound like a dictionary.

4. Location & profile: Google Gemini knows (if allowed) where you are. So the recommendation for a restaurant will be customized locally.

Personalization:
Models learn from their chat history (e.g. preferences).
Different users have different contexts - a prompt with you (experienced) might go deeper than with a newbie. Examples: "Tell a story" - child-friendly for a child, philosophical for you.
Adapt prompts: "Adapt the answer to a 40-year-old with basic AI knowledge."

Rhetorical questions and decision-making aids

• Should I ask a question in Google or in an AI like ChatGPT? Use Google for pure facts (e.g., "Weather today"), AI for synthesis (e.g., "How does climate affect travel?"). Recognize: If you need explanations, AI; for verification, search engine.

• Can AI be creative? Yes, but based on patterns - use it as an idea starter.

Hallucinations: Unmasking the hoax

Why does AI hallucinate?

Imagine you have to continue a story but you don't know the ending. You would try to find an ending that "sounds right". This is exactly what the AI does. It wants to please you and would rather give you the wrong answer than "I don't know" (although that gets better).

How do I check the sources?

1. Counter question: "Tell me the specific website or book you got this information from."

2. Perplexity check: Copy the claim into Perplexity. As this tool searches the web, it will immediately show you whether there is any real evidence.

3. Logic test: AIs are bad at arithmetic and logic puzzles. Always check numbers

The art of "prompting" (writing instructions)

A good prompt needs structure. Use the R-A-F formula:

• Role: "You are an experienced IT expert for seniors."

• Task: "Explain to me how to create a secure password."

• Format: "Write it as a short checklist with a maximum of 5 points."

My first pro prompt

Use this space to draft your first structured prompt during the presentation.

My role for the AI:

(e.g. tour guide, nutritionist, grandchild coach)

My exact task:

(e.g. Create a list, write an email, summarize a topic)

Restrictions / format:

(e.g. maximum 100 words, in tabular form, in polite "you" form)