Key question
What do people actually mean when they say AI?
Free field guide
2026 AI Guide
A one-page curriculum for understanding AI from the top down: the core hierarchy, modern model capabilities, and the OpenAI platform terms teams now need for GPT-5.5, Responses API, agents, tools, and realtime AI.
The 2026 AI Guide supports corporate AI training in Switzerland — from foundation models and Generative AI to Agentic AI, MCP, and governance patterns used in AI Workshop programs.
12
Core concepts
GPT-5.5
OpenAI model language
Responses
Modern API vocabulary
1 Layer 1 AI Overview AI is the umbrella field. Everything below sits inside it. 2 Layer 2 Machine Learning Machine Learning is a child of AI. It learns patterns from data instead of relying only on fixed rules. 3 Layer 3 Deep Learning Deep Learning is a child of Machine Learning. It uses many neural-network layers to learn complex features. 4 Layer 4 Foundation Models Foundation Models are a modern child of Deep Learning. They become reusable engines for many downstream tasks.
Chapter 1
The AI stack
Start with the umbrella idea of AI, then narrow into the child layers that make modern systems possible.
By the end of this chapter
You should understand the nested relationship between AI, Machine Learning, Deep Learning, Neural Networks, and Foundation Models.
Chapter 1 / Lesson 01 Foundation Updated November 27, 2025
AI Overview
From Turing machines to Transformer models.
Where it sits in the stack
AI Overview
What to remember
AI is the broad field. Everything else in this guide is a more specific layer inside it.
Artificial Intelligence is the broad discipline of creating intelligent machines. It is the umbrella term that encompasses everything from simple rule-based systems to the complex Large Language Models of today.
Concept map +
How is AI structured? What are the types of AI? Why has AI exploded recently? A Brief History of AI
Theory 01
How is AI structured?
To understand AI, visualize concentric circles. The outermost circle is Artificial Intelligence-the grand vision. Inside that is Machine Learning-the technique of learning from data. Inside that is Deep Learning-using neural networks. And at the cutting edge, we find Generative AI-models that create.
Theory 02
What are the types of AI?
We classify AI capability into three stages:
- ANI (Artificial Narrow Intelligence): AI that excels at one specific task (e.g., playing Chess, recommending movies). This is where we are today.
- AGI (Artificial General Intelligence): AI that possesses the ability to understand, learn, and apply knowledge across a wide variety of tasks, matching human capability.
- ASI (Artificial Super Intelligence): An intellect that is much smarter than the best human brains in practically every field.
Theory 03
Why has AI exploded recently?
It is the convergence of three factors:
- Big Data: The internet provided the fuel.
- Compute Power: GPUs provided the engine.
- Better Algorithms: Transformers provided the map.
Theory 04
A Brief History of AI
- 1950: Alan Turing proposes the Turing Test.
- 1956: The term "Artificial Intelligence" is coined at Dartmouth.
- 1997: Deep Blue beats Garry Kasparov at Chess.
- 2012: AlexNet revolutionizes computer vision (Deep Learning boom).
- 2017: The "Attention Is All You Need" paper introduces Transformers.
- 2022: ChatGPT is released, bringing Generative AI to the masses.
Example: A Brief History of AI +
- 1950: Alan Turing proposes the Turing Test.
- 1956: The term "Artificial Intelligence" is coined at Dartmouth.
- 1997: Deep Blue beats Garry Kasparov at Chess.
- 2012: AlexNet revolutionizes computer vision (Deep Learning boom).
- 2017: The "Attention Is All You Need" paper introduces Transformers.
- 2022: ChatGPT is released, bringing Generative AI to the masses.
Common question +
Are we close to AGI?
Estimates vary wildly from a few years to a few decades. The rapid progress of LLMs has accelerated these timelines, but significant hurdles in reasoning and physical world understanding remain.
Chapter 1 / Lesson 02 Core Concept Updated November 27, 2025
Machine Learning
Computers that learn from data and improve with experience.
Where it sits in the stack
AI Overview -> Machine Learning
Key question
How does a machine get better without being reprogrammed for every case?
What to remember
Machine Learning is the learning engine inside AI, powered by data rather than explicit instructions for every outcome.
Machine Learning is a subset of Artificial Intelligence (AI) where computers learn from data and improve through experience without being explicitly programmed. Algorithms are trained to find patterns and correlations in large datasets to make the best decisions and predictions. With practice and more data, these applications become increasingly accurate.
Concept map +
AI, Machine Learning, and Deep Learning Neural Networks and Deep Learning The 4 Types of Machine Learning Real-World Applications Challenges in Machine Learning
Theory 01
AI, Machine Learning, and Deep Learning
Think of them as concentric circles. Artificial Intelligence is the broad discipline. Machine Learning is a subset within AI that allows machines to learn from data. Inside Machine Learning is Deep Learning, and within that are Artificial Neural Networks. AI processes data to make decisions; ML algorithms allow AI to learn and get smarter without additional programming.
Theory 02
Neural Networks and Deep Learning
Artificial Neural Networks mimic the biological brain, with nodes (neurons) grouped in layers working in parallel. They strengthen connections to improve pattern recognition.
Deep Learning involves many layers of these networks and huge volumes of complex data. It extracts features hierarchically: a system might recognize a plant in the first layer, a flower in the next, and a yellow daisy in the last.
Theory 03
The 4 Types of Machine Learning
- Supervised Learning: Learning by example. The machine is given labeled inputs and outputs (e.g., images of daisies labeled "daisy"). It learns to map new inputs to the correct output.
- Unsupervised Learning: No answer key. The machine analyzes unlabeled data to find hidden patterns, clusters, or structures, similar to how humans observe and categorize the world.
- Semi-Supervised Learning: Uses a small amount of labeled data to guide the analysis of a large amount of unlabeled data. This speeds up learning and improves accuracy.
- Reinforcement Learning: Learning by trial and error. The system receives "rewards" for good actions and "penalties" for bad ones, optimizing its strategy over time (e.g., playing chess).
Theory 04
Real-World Applications
- Recommendation Engines: Streaming services (Netflix, Spotify) analyzing viewing habits to suggest content.
- Dynamic Marketing: Analyzing customer data to personalize marketing and engage in real-time.
- ERP & Automation: Optimizing workflows and automating repetitive tasks using business data.
- Predictive Maintenance: IoT sensors on machinery predicting failures before they happen, saving costs and preventing downtime.
Theory 05
Challenges in Machine Learning
Bias and Spurious Correlations: Models can learn incorrect associations (e.g., correlating margarine consumption with divorce rates) if the data is flawed or if they find coincidental patterns.
The Black Box Problem: Complex models like deep neural networks can be difficult to interpret. It is often unclear how or why a specific decision was made, which poses risks in critical fields.
Example: Real-World Applications +
- Recommendation Engines: Streaming services (Netflix, Spotify) analyzing viewing habits to suggest content.
- Dynamic Marketing: Analyzing customer data to personalize marketing and engage in real-time.
- ERP & Automation: Optimizing workflows and automating repetitive tasks using business data.
- Predictive Maintenance: IoT sensors on machinery predicting failures before they happen, saving costs and preventing downtime.
Common question +
What is the difference between AI and Machine Learning?
AI is the broader concept of machines acting intelligently. Machine Learning is a specific subset of AI where machines learn from data to improve their performance without being explicitly programmed for every task.
Chapter 1 / Lesson 03 Advanced Updated June 13, 2026
Deep Learning
Unlocking the power of high-dimensional data.
Where it sits in the stack
AI Overview -> Machine Learning -> Deep Learning
Key question
Why did modern AI become dramatically more capable?
What to remember
Deep Learning is scaled Machine Learning built on many layers, which makes it strong on high-dimensional and unstructured data.
Deep Learning is a specialized subset of Machine Learning inspired by the structure of the human brain. It uses multi-layered neural networks to learn from vast amounts of unstructured data like images, audio, and text.
Concept map +
Why is it called "Deep" Learning? What is Automatic Feature Extraction? Why is Deep Learning important now? Key Architectures
Theory 01
Why is it called "Deep" Learning?
The "Deep" in Deep Learning refers to the number of layers in the neural network. Traditional neural networks might have 2-3 layers. Deep learning models can have hundreds or thousands. This depth allows the model to learn a hierarchy of features-from simple edges and textures to complex shapes and objects.
Theory 02
What is Automatic Feature Extraction?
In traditional ML, humans had to manually select features (e.g., "does this image have ears?"). In Deep Learning, the network performs automatic feature extraction. It learns what features are important directly from the raw pixels or text.
Theory 03
Why is Deep Learning important now?
Deep Learning is the technology behind self-driving cars, voice assistants, facial recognition, and the recent generative AI boom. It thrives on scale-more data and more compute usually lead to better performance.
Theory 04
Key Architectures
- CNNs (Convolutional Neural Networks): The kings of computer vision.
- RNNs (Recurrent Neural Networks): Good for time-series and sequential data.
- Transformers: The state-of-the-art for natural language processing (NLP).
Example: Why is Deep Learning important now? +
Deep Learning is the technology behind self-driving cars, voice assistants, facial recognition, and the recent generative AI boom. It thrives on scale-more data and more compute usually lead to better performance.
Common question +
Is Deep Learning the same as Neural Networks?
Deep Learning is essentially the use of *deep* neural networks. So all Deep Learning involves neural networks, but not all neural networks are 'deep' (though in modern context, the terms are often used interchangeably).
Chapter 1 / Lesson 04 Technical Updated November 27, 2025
Neural Networks
The mathematical architecture of the mind.
Where it sits in the stack
AI Overview -> Machine Learning -> Deep Learning -> Neural Networks
Key question
What is the actual mechanism doing the learning inside Deep Learning?
What to remember
Neural Networks are the mathematical structure that powers Deep Learning and makes modern model scaling possible.
Artificial Neural Networks (ANNs) are computing systems vaguely inspired by the biological neural networks that constitute animal brains. They are the fundamental building blocks of Deep Learning.
Concept map +
How does an Artificial Neuron work? What are the main types of Neural Networks? How do Neural Networks learn?
Theory 01
How does an Artificial Neuron work?
A neuron takes multiple inputs, multiplies them by weights (importance), adds a bias (threshold), and passes the result through an activation function (non-linearity). If the signal is strong enough, the neuron "fires" and passes information to the next layer.
Theory 02
What are the main types of Neural Networks?
- Feedforward NN: The simplest type. Information moves in one direction.
- CNN (Convolutional Neural Network): Specialized for processing grid-like data (images). It scans the image with filters to detect patterns.
- RNN (Recurrent Neural Network): Designed for sequential data (time series, text). It has a "memory" of previous inputs.
- Transformer: The modern architecture for language. It uses "attention" mechanisms to weigh the importance of different parts of the input data simultaneously.
Theory 03
How do Neural Networks learn?
They learn through a process called Backpropagation. The network makes a guess, compares it to the actual answer to calculate the loss (error), and then works backward to adjust the weights to minimize that error. This is repeated millions of times.
Example: What is an activation function? +
It's a mathematical function (like ReLU or Sigmoid) attached to each neuron that decides whether it should be activated. It introduces non-linearity, allowing the network to learn complex patterns.
Common question +
What is the 'Black Box' problem?
Neural networks can be so complex that even their creators don't fully understand how they arrive at a specific decision. This lack of interpretability is a major challenge in high-stakes fields like medicine.
Chapter 1 / Lesson 05 Modern AI Updated November 27, 2025
Foundation Models
One model, infinite applications.
Where it sits in the stack
AI Overview -> Machine Learning -> Deep Learning -> Foundation Models
Key question
Why do a few large models now power so many different tasks?
What to remember
Foundation Models are general-purpose engines trained at massive scale and adapted to many downstream uses.
A Foundation Model is a large-scale AI model trained on broad data that can be adapted to many downstream tasks. Modern foundation models are increasingly multimodal, tool-capable, and used as reasoning engines inside assistants, APIs, agents, and enterprise workflows.
Concept map +
What makes a model a "Foundation Model"? What is "Emergence"? How are they built? Leading foundation model families
Theory 01
What makes a model a "Foundation Model"?
It must be broadly capable. Unlike previous models designed for one narrow task, a foundation model can draft, analyze, code, translate, summarize, interpret images, use tools, and adapt to many product experiences through prompting, retrieval, fine-tuning, or orchestration.
Theory 02
What is "Emergence"?
Foundation models exhibit emergence-capabilities that were not explicitly trained for. For example, a model trained simply to predict the next word in a sentence might emerge with the ability to translate languages, write code, or solve logic puzzles.
Theory 03
How are they built?
The lifecycle involves two stages:
- Pre-training: The expensive, compute-intensive phase where the model learns general patterns from massive datasets (e.g., "learning to read and write").
- Fine-tuning: The adaptation phase where the model is specialized for a specific task or behavior (e.g., "learning to be a helpful assistant").
Theory 04
Leading foundation model families
Prominent model families include GPT-5.5 and other GPT models from OpenAI, Claude from Anthropic, Gemini from Google DeepMind, Llama from Meta, Mistral models, and specialist image, audio, and realtime models. The important point is no longer only the model name, but the surrounding platform: tools, state, retrieval, structured outputs, governance, and agent orchestration.
Example: Leading foundation model families +
Prominent model families include GPT-5.5 and other GPT models from OpenAI, Claude from Anthropic, Gemini from Google DeepMind, Llama from Meta, Mistral models, and specialist image, audio, and realtime models. The important point is no longer only the model name, but the surrounding platform: tools, state, retrieval, structured outputs, governance, and agent orchestration.
Common question +
Are Foundation Models the same as LLMs?
LLMs (Large Language Models) are a *type* of foundation model focused on text. But foundation models can also be multimodal, handling images, audio, and video.
Chapter 2
From models to generated output
Once the stack is clear, you can understand how modern systems generate content and who is shaping that layer.
By the end of this chapter
You should be able to explain what Generative AI is, why it depends on foundation models, and how major model providers differ.
Chapter 2 / Lesson 06 Creative AI Updated November 27, 2025
Generative AI
From analysis to creation.
Where it sits in the stack
AI Overview -> Machine Learning -> Deep Learning -> Foundation Models -> Generative AI
Key question
How does modern AI move from analysis into creation?
What to remember
Generative AI is the creation layer. It uses modern models to produce new text, images, code, audio, and more.
Generative AI refers to algorithms that can create new content-including audio, code, images, text, simulations, and videos. Unlike traditional AI which classifies or predicts, Generative AI produces novel outputs.
Concept map +
Discriminative vs. Generative AI How does Generative AI work? The Creative Revolution Use Cases
Theory 01
Discriminative vs. Generative AI
Traditional AI is Discriminative: It draws a line to separate data (e.g., "Is this a cat or a dog?"). Generative AI is Creative: It learns the distribution of the data to create new examples (e.g., "Draw me a cat that never existed").
Theory 02
How does Generative AI work?
Generative models, such as Diffusion Models (for images) or Transformers (for text), learn the underlying structure of the training data. They then use probability to assemble new patterns that follow those structures but are not identical copies.
Theory 03
The Creative Revolution
Generative AI is democratizing creativity. It acts as a co-pilot for writers, artists, coders, and designers, allowing them to iterate faster and explore new ideas. It is shifting the bottleneck from "skill" to "imagination".
Theory 04
Use Cases
- Marketing: Generating ad copy and visuals.
- Coding: Writing boilerplate code and documentation.
- Entertainment: Creating game assets and scripts.
- Science: Generating novel protein structures for drug discovery.
Example: Use Cases +
- Marketing: Generating ad copy and visuals.
- Coding: Writing boilerplate code and documentation.
- Entertainment: Creating game assets and scripts.
- Science: Generating novel protein structures for drug discovery.
Common question +
Does Generative AI steal art?
This is a complex legal and ethical issue. Models learn from existing data, but they don't 'copy-paste'. They learn styles and concepts. However, the rights of the original creators of the training data are a subject of active debate and litigation.
Chapter 2 / Lesson 07 Industry Updated June 13, 2026
LLM Players
The model providers, platforms, and product ecosystems shaping applied AI.
Where it sits in the stack
AI Overview -> Machine Learning -> Deep Learning -> Foundation Models -> LLM Players
Key question
Who is shaping the model layer that everyone builds on?
What to remember
Model providers are not interchangeable. Each one makes different tradeoffs in openness, performance, integration, and control.
The Large Language Model landscape has moved beyond simple chatbot comparison. Teams now compare model quality, reasoning capability, tool use, privacy controls, multimodal support, agent orchestration, and ecosystem fit.
Concept map +
Who are the key players now? What should teams compare?
Theory 01
Who are the key players now?
- OpenAI: GPT-5.5, ChatGPT, the Responses API, hosted tools, Realtime API, Codex, and the Agents SDK make OpenAI a full application and agent platform, not just a model provider.
- Google DeepMind: Gemini brings strong multimodal capabilities and deep integration with the Google ecosystem.
- Anthropic: Claude is widely used for writing, analysis, coding, and long-context knowledge work, with a strong safety positioning.
- Meta: Llama and open-weight models remain important for organizations that want more control over deployment, privacy, and cost.
- Mistral and European providers: Mistral and other European players matter for sovereignty, efficiency, and enterprise procurement in Switzerland and the EU.
Theory 02
What should teams compare?
- Model quality: reasoning, coding, writing, multilingual performance, and multimodal understanding.
- Application layer: APIs, tools, connectors, state handling, structured outputs, and agent orchestration.
- Deployment model: API, enterprise workspace, cloud tenant, open weights, or local deployment.
- Governance: data retention, auditability, safety controls, human review, and procurement fit.
Example: What should teams compare? +
- Model quality: reasoning, coding, writing, multilingual performance, and multimodal understanding.
- Application layer: APIs, tools, connectors, state handling, structured outputs, and agent orchestration.
- Deployment model: API, enterprise workspace, cloud tenant, open weights, or local deployment.
- Governance: data retention, auditability, safety controls, human review, and procurement fit.
Common question +
Which model is the best?
It depends on the workflow. GPT-5.5 is a strong OpenAI choice for production work that needs reasoning, tool use, coding, and polished output. Claude, Gemini, Llama, Mistral, and other models may fit better when the priority is context handling, ecosystem integration, local control, price, or procurement requirements.
Chapter 3
From models to products, tools, and agents
After the model layer, the next question is how people actually use AI in real work: tools, assistants, agents, realtime experiences, and hands-on practice.
By the end of this chapter
You should understand how AI becomes tools, how to compare products, and why current OpenAI language now includes hosted tools, tool search, MCP, Agents SDK, and Realtime sessions.
Chapter 3 / Lesson 08 Practical Updated June 13, 2026
AI Tools
From chat interfaces to agentic workflows.
Where it sits in the stack
AI Overview -> Machine Learning -> Deep Learning -> Foundation Models -> Generative AI -> AI Tools
Key question
How do models become useful products in everyday work?
What to remember
AI tools are the interface layer between advanced models and real human tasks.
AI tools are the practical interface between models and work. In 2026, the important shift is from standalone chat prompts to governed workflows with retrieval, structured outputs, hosted tools, MCP connectors, realtime voice, and human approval points.
Concept map +
What are the main categories of AI tools? How to choose the right tool?
Theory 01
What are the main categories of AI tools?
- Chat and reasoning assistants: ChatGPT, Claude, Gemini, and Copilot help with drafting, analysis, planning, and decision support.
- Retrieval and research tools: Web search, file search, RAG systems, and enterprise search ground answers in current or private sources.
- Creative generation: GPT Image, Midjourney, Runway, Firefly, and Canva support image, video, and design workflows.
- Coding and software agents: Codex, GitHub Copilot, Cursor, and related tools help navigate codebases, write changes, review diffs, and run tests.
- Realtime and audio tools: Voice agents, transcription, translation, and speech generation support live customer and team workflows.
Theory 02
How to choose the right tool?
- Define the workflow: writing, analysis, customer support, HR, finance, coding, research, or voice.
- Check the model and API layer: Does the tool support reasoning, multimodal input, structured output, retrieval, tools, or agent workflows?
- Check privacy and governance: What data is stored, used for training, logged, or exported? Who can audit actions?
- Look for integration: Does it connect to your documents, CRM, Microsoft 365, Google Workspace, ticketing, or internal APIs?
- Evaluate with real tasks: Compare accuracy, speed, cost, and human-review needs on representative examples.
Example: How to choose the right tool? +
- Define the workflow: writing, analysis, customer support, HR, finance, coding, research, or voice.
- Check the model and API layer: Does the tool support reasoning, multimodal input, structured output, retrieval, tools, or agent workflows?
- Check privacy and governance: What data is stored, used for training, logged, or exported? Who can audit actions?
- Look for integration: Does it connect to your documents, CRM, Microsoft 365, Google Workspace, ticketing, or internal APIs?
- Evaluate with real tasks: Compare accuracy, speed, cost, and human-review needs on representative examples.
Common question +
Will these tools replace my job?
AI is more likely to change workflows before it fully replaces roles. The practical skill is knowing how to delegate, verify, and govern AI-assisted work.
Chapter 3 / Lesson 09 Resources Updated June 13, 2026
AI Tools Directory
A curated map of current AI product categories.
Where it sits in the stack
AI Overview -> Machine Learning -> Deep Learning -> Foundation Models -> Generative AI -> AI Tools -> AI Tools Directory
Key question
How do you evaluate the crowded tool market without getting lost?
What to remember
The right tool choice depends on workflow, privacy, integration, and the model sitting underneath the interface.
Navigating AI tools is easier when you organize them by job to be done: assistant, search, agent, code, creative, voice, automation, and enterprise knowledge.
Concept map +
Chat & Assistants OpenAI platform concepts Development and agentic work Creative and audio
Theory 01
Chat & Assistants
- ChatGPT (OpenAI): Conversational AI, reasoning, multimodal work, projects, enterprise usage, and custom workflows.
- Claude (Anthropic): Strong writing, analysis, coding, and long-context knowledge work.
- Gemini (Google): Multimodal assistant integrated with Google apps.
- Microsoft Copilot: AI inside Microsoft 365, Teams, Outlook, Word, Excel, PowerPoint, and enterprise workflows.
- Perplexity: AI-powered search and answer engine.
Theory 02
OpenAI platform concepts
- Responses API: model responses, tools, multimodal input, and state.
- Hosted tools: web search, file search, code interpreter, image generation, and computer use.
- Tool search: defer tool definitions and load only what is relevant.
- Agents SDK: orchestration, handoffs, tracing, guardrails, and human review.
- Realtime API: voice agents, translation, and live transcription.
Theory 03
Development and agentic work
- Codex: OpenAI coding agent for software development tasks.
- Cursor: AI-first code editor.
- GitHub Copilot: Widely used coding assistant in developer environments.
- Vercel v0: Generative UI and frontend prototyping.
- n8n and Zapier: Workflow automation where AI can be added to business processes.
Theory 04
Creative and audio
- GPT Image: OpenAI image generation and editing.
- Midjourney: High-quality artistic image generation.
- Runway: AI video generation and editing.
- Firefly: Adobe creative AI inside design workflows.
- Realtime voice and transcription tools: live assistant, interpreter, and transcript workflows.
Example: Chat & Assistants +
- ChatGPT (OpenAI): Conversational AI, reasoning, multimodal work, projects, enterprise usage, and custom workflows.
- Claude (Anthropic): Strong writing, analysis, coding, and long-context knowledge work.
- Gemini (Google): Multimodal assistant integrated with Google apps.
- Microsoft Copilot: AI inside Microsoft 365, Teams, Outlook, Word, Excel, PowerPoint, and enterprise workflows.
- Perplexity: AI-powered search and answer engine.
Chapter 3 / Lesson 10 Practice Updated June 13, 2026
Interactive Exercises
Learn by doing.
Where it sits in the stack
AI Overview -> Machine Learning -> Interactive Exercises
Key question
How do abstract AI ideas become intuition instead of memorized jargon?
What to remember
Interactive learning is the shortest path from theory to intuition.
Theory is essential, but practice makes perfect. These interactive simulations and games are designed to build your intuition for how AI systems actually work.
Concept map +
Why Interactive Learning? Available Modules
Theory 01
Why Interactive Learning?
AI concepts like "Gradient Descent" or "Backpropagation" can be abstract and mathematical. Interactive visualizations allow you to see the math in action, building a deeper, intuitive understanding.
Theory 02
Available Modules
- Neuron Visualizer: See how inputs, weights, and biases combine to fire a neuron.
- Network Playground: Build and train simple neural networks in your browser.
- Gradient Descent Sim: Visualize how models minimize error by descending a loss landscape.
- Hyperparameter Sandbox: Experiment with learning rates and batch sizes to see their effect on training.
Example: Available Modules +
- Neuron Visualizer: See how inputs, weights, and biases combine to fire a neuron.
- Network Playground: Build and train simple neural networks in your browser.
- Gradient Descent Sim: Visualize how models minimize error by descending a loss landscape.
- Hyperparameter Sandbox: Experiment with learning rates and batch sizes to see their effect on training.
Chapter 4
Human judgment and shared language
Finish by reconnecting AI to people: human intelligence, augmented decision-making, and the vocabulary needed for clear conversations.
By the end of this chapter
You should be able to discuss AI with more precision, better judgment, and a shared vocabulary across teams.
Chapter 4 / Lesson 11 Fundamentals Updated November 27, 2025
Intelligence
IQ, EQ, and AI shape our decision-making.
Where it sits in the stack
AI Overview -> Intelligence
Key question
What still belongs to humans in an AI-shaped world?
What to remember
The strongest future is not human or machine alone. It is augmented intelligence with clear human oversight.
Understanding how human intelligence, artificial intelligence, and augmented intelligence complement each other is key to navigating the future. Intelligence is not just about processing power; it's about the synergy between biological and synthetic cognition.
Concept map +
Three Types of Intelligence Artificial Intelligence (AI) Emotional Intelligence (EQ)
Theory 01
Three Types of Intelligence
We can categorize intelligence into three distinct forms that interact in the modern world:
| Feature | Humans | Machines | Augmented Intelligence |
|---|---|---|---|
| Data Handling | Understand and generalize concepts | Process and analyze large volumes of data | Combine context with data-driven insights |
| Repetition | Prone to fatigue | Perform repetitive tasks with high accuracy | Automate tasks while preserving human oversight |
| Creativity | Flexible problem-solving | Limited creative capacity | Enhance human creativity with smart tools |
| Emotional Insight | Empathy & customer care | No emotional understanding | Human-led empathy, supported by smart assistance |
Theory 02
Artificial Intelligence (AI)
Definition: The ability of machines to think and reflect like humans, attempting to replicate human intelligence with machines.
Capabilities:
- Reasoning: Logical thinking and inference.
- Natural Communication: Understanding and generating human language.
- Problem-solving: Finding solutions to complex challenges.
Characteristics:
- Replaces Human Effort: Automates tasks traditionally done by humans.
- Performs Tasks Independently: Operates without constant human intervention.
Theory 03
Emotional Intelligence (EQ)
"Why aren't we more compassionate?" - Daniel Goleman
Emotional Intelligence operates through the basal ganglia (the wisdom center of the brain). It guides decisions based on emotional valence (what felt good/bad). Unlike the neocortex, it does not speak in words but is connected to emotions and the gut.
Key Insight: Combine EQ + IQ for better decisions.
Example: What is Augmented Intelligence? +
Augmented Intelligence is a design pattern for a human-centered partnership model of people and AI working together to enhance cognitive performance, including learning, decision making, and new experiences.
Common question +
What is Augmented Intelligence?
Augmented Intelligence is a design pattern for a human-centered partnership model of people and AI working together to enhance cognitive performance, including learning, decision making, and new experiences.
Chapter 4 / Lesson 12 Dictionary Updated November 27, 2025
AI Concepts & Terminology
Speak the language of the future.
Where it sits in the stack
AI Overview -> AI Concepts & Terminology
Key question
Which terms should everyone use consistently after reading this guide?
What to remember
A strong AI guide should end with alignment: people leaving with the same words for the same concepts.
The field of AI is filled with jargon. This dictionary provides clear, concise definitions for the most important terms you need to know.
Concept map +
A-E F-L M-Z
Theory 01
A-E
Agent: A system that can plan, use tools, manage state, and execute multi-step work toward a goal.
Algorithm: A set of rules, calculations, or procedures used by software or models to process information.
Alignment: The work of making AI behavior match human intent, values, and product expectations.
Bias: Systematic errors in data, design, deployment, or outputs that can unfairly affect results.
Computer use: A tool pattern where a model can interact with a computer interface to complete UI-based tasks.
Theory 02
F-L
File search: A retrieval tool that lets a model search uploaded or connected files for relevant context.
Fine-tuning: The process of adapting a pre-trained model on a targeted dataset for a more specific behavior or domain.
Function calling: A way to expose custom software actions to a model with typed inputs and predictable outputs.
Hallucination: When an AI generates incorrect or unsupported information confidently.
Hosted tools: OpenAI-provided tools such as web search, file search, code interpreter, image generation, and computer use.
LLM (Large Language Model): A model specialized in understanding and generating language.
Theory 03
M-Z
MCP (Model Context Protocol): A protocol for connecting models and agents to external tools, systems, and data sources.
Multimodal: AI that can work across multiple input or output types, such as text, images, audio, video, and structured data.
Realtime API: OpenAI sessions for low-latency voice agents, translation, transcription, and live multimodal experiences.
Reasoning effort: A model setting that controls how much reasoning budget is used for supported models.
Responses API: The OpenAI API surface for model responses, tools, multimodal input, state, and agentic workflows.
Structured Outputs: A way to make model output follow a defined schema for reliable downstream use.
Token: A unit of text or data processed by a model.
Tool search: A pattern for loading only the relevant tool definitions when a tool catalog is large.
Example: F-L +
File search: A retrieval tool that lets a model search uploaded or connected files for relevant context.
Fine-tuning: The process of adapting a pre-trained model on a targeted dataset for a more specific behavior or domain.
Function calling: A way to expose custom software actions to a model with typed inputs and predictable outputs.
Hallucination: When an AI generates incorrect or unsupported information confidently.
Hosted tools: OpenAI-provided tools such as web search, file search, code interpreter, image generation, and computer use.
LLM (Large Language Model): A model specialized in understanding and generating language.
Extended guide
Modern AI, in plain language
These are the extra concepts from the workshop guide that help people connect the model stack to what they actually see in products today.
LLM in plain language The language-specialized child of the model stack. +
Definition
What it is
A Large Language Model is a very large model specialized in understanding and generating language. In practice, it acts like a text engine that can summarize, draft, explain, translate, and reason over documents.
Analogy
Simple image
Think of someone who has read a vast library and can help you write, explain, or compare ideas on demand.
Examples
Tools and providers
- OpenAI: ChatGPT, GPT-5.5, Responses API, Realtime API, Agents SDK
- Anthropic and Claude
- Google DeepMind and Gemini
- Meta and Llama
- Mistral, Cohere, xAI, Aleph Alpha, and DeepSeek
Responses API in plain language The application layer for modern OpenAI workflows. +
Definition
What it is
The Responses API is the OpenAI API surface designed for model responses, multimodal input, tool use, multi-turn state, and agentic workflows. For teams building with current OpenAI models, it is the vocabulary to learn before older chat-only patterns.
Controls
What builders configure
- reasoning.effort controls how much reasoning budget the model uses.
- text.verbosity controls how concise or detailed the answer should be.
- Structured Outputs validate JSON or schema-shaped answers without relying only on prompt wording.
- previous_response_id handles multi-turn state when the application keeps conversation context.
Multimodal in plain language Models that do more than just text. +
Definition
What it is
A multimodal system can understand more than one type of input, such as text, images, audio, or video, and combine them in one answer.
Example
Simple example
You upload a photo of a chart and ask a question in text. The model uses both the image and the language prompt to answer.
AI agents in plain language When a model starts planning and acting across steps. +
Definition
What it is
An AI agent is a system that can plan steps, use tools, execute tasks, and sometimes coordinate with other agents to reach a goal. In OpenAI's current platform language, the Agents SDK adds patterns for orchestration, tracing, handoffs, guardrails, human review, and state management.
Example
Simple workflow
- Research a topic
- Summarize the useful points
- Organize a presentation outline
- Draft a first version
Attention
What to watch
- Verify sources for important topics
- Define a clear objective and instructions
- Use approvals for risky side effects
- Trace tool calls and handoffs when agents affect business workflows
- Review outputs before sharing them
Tools, MCP, and hosted capabilities How modern AI systems connect to real work. +
Hosted tools
OpenAI platform tools to know
- Web search for current public information
- File search for retrieval over uploaded documents
- Code interpreter for data work and calculations
- Image generation with GPT Image
- Computer use for workflows that need UI interaction
Tool access
How tools are exposed
- Function calling exposes your own business logic with typed inputs.
- Tool search loads only the relevant tool definitions when a catalog is large.
- Remote MCP servers connect models to external tools and data sources.
- ChatGPT Apps use an MCP server plus a web component UI to create interactive apps inside ChatGPT.
Realtime AI in plain language Low-latency voice, translation, and live transcription. +
Definition
What it is
Realtime sessions keep a live connection open while an app streams audio, receives events, calls tools, and updates state. Use them when a product needs interactive voice or multimodal behavior, not just a single request and response.
Examples
Common architectures
- Voice-agent sessions for speech-to-speech assistants
- Realtime translation sessions for live interpretation
- Realtime transcription sessions for streaming transcript deltas
- WebRTC for browser and mobile audio, WebSockets for server media pipelines
Practical guardrails
Safety, common sense, and shared language
Simple reflexes
- Do not paste passwords, banking details, or sensitive documents into consumer tools.
- Verify important outputs in health, legal, finance, and compliance contexts.
- Remember that AI can confidently invent details.
- Use enterprise controls, approved tools, and human review when AI touches customer, HR, legal, or financial data.
- For agent workflows, separate tool access from approval rules.
- Use AI as an assistant, not as absolute truth.
Myth vs reality
- Myth: if it sounds polished, it must be true. Reality: well-written output can still be wrong.
- Myth: AI always replaces the human. Reality: it works best when the human sets the frame and reviews the result.
30-second version
- AI means machines imitating some human capabilities.
- Machine Learning means they improve from examples.
- Neural Networks means they learn patterns through layers.
- Deep Learning means very large, multi-layer neural systems.
- LLMs are large language-focused models.
- Generative AI creates new content.
- The Responses API is the modern OpenAI layer for model responses, tools, and state.
- Reasoning effort, verbosity, Structured Outputs, hosted tools, and prompt caching are core production concepts.
- Agents plan and execute multi-step tasks through tools, handoffs, guardrails, and tracing.
- Realtime sessions power low-latency voice, translation, and transcription.
5-minute challenge
- Ask an LLM to explain the difference between AI, ML, and LLM for a beginner.
- Then ask for one example from your own work or daily life.
- Finally, verify one important claim with a reliable source.
Visual glossary
Shared language people can actually remember
Artificial Intelligence
A machine doing tasks that normally require human intelligence.
Use: Used to understand, decide, recommend, and create.
Example: Spam filters, translation, recommendations.
Machine Learning
The part of AI that learns from examples instead of following only fixed rules.
Use: Used to predict, classify, and segment.
Example: Spam or not spam, price estimation.
Language Model
A large model specialized in understanding and generating text.
Use: Used to summarize, draft, explain, and translate.
Example: ChatGPT, Claude, Gemini.
Current OpenAI model family
A current OpenAI model family designed for strong reasoning, tool-heavy agents, coding, grounded assistants, and polished customer-facing workflows.
Use: Used when quality, execution, and reliability matter more than the fastest possible turn.
Example: A customer support agent that searches, reasons, drafts, and cites evidence.
Responses API
The OpenAI API layer for model responses, multimodal input, tools, state, and agentic workflows.
Use: Used to build current OpenAI applications instead of relying only on older chat-only patterns.
Example: A workflow that uses file search, structured output, and previous_response_id.
reasoning.effort
A setting that controls how much reasoning budget a supported model uses.
Use: Use low or medium for many production workflows, and increase only when evals justify the cost and latency.
Example: Set higher effort for complex legal review, lower effort for simple classification.
Structured Outputs
A way to make model output follow a JSON schema or exact structure with stronger validation.
Use: Use it when downstream software depends on reliable fields, not prose.
Example: Return a CRM lead object with company, role, budget, urgency, and next action.
Plan + Tools + Execution
A system that plans steps, uses tools, and executes a task with state, review rules, and traceable actions.
Use: Used to automate multi-step workflows where a model needs tools or handoffs.
Example: Research + file search + spreadsheet analysis + summary + approval request.
Tool Protocol
A standard for connecting models to tools and external data sources.
Use: Lets AI fetch context or act through tools.
Example: A remote MCP server that exposes CRM, calendar, or document tools.
Live AI sessions
A live connection for low-latency audio, translation, transcription, and voice-agent experiences.
Use: Used when the user experience depends on streaming events rather than a single request.
Example: A browser voice agent using WebRTC and server-side tools.
Keep these five lines in your head
- AI is the big umbrella.
- Machine Learning means learning from examples.
- LLM means language-focused model; reasoning models add stronger multi-step problem solving.
- Responses API is the modern OpenAI application layer for model output, tools, and state.
- Structured Outputs are for reliable data shapes; prompting alone is not enough.
- Agent means planning plus tools plus execution, with guardrails and tracing.
- MCP connects models to external tools and data sources.