AI FLUENCY · PHASE 1: AWARENESS
Module 3 — The Four Properties of AI | Lesson 012
Steerability: Why Does AI Follow Instructions?
Learn how instructions guide AI responses and why the same question can produce different explanations.
Watch this 54-second silent visual lesson. Then explore the explanation and examples below.
Watch on YouTube →Introduction
Have you noticed that ChatGPT, Claude, or Gemini can explain the same topic in different ways depending on your instructions?
You can ask for a simple explanation, a professional summary, a creative story, or a step-by-step guide.
The model can often adapt its response to your needs.
This capability is called steerability.
Steerability is the fourth property explored in Module 3 of AI Fluency.
1. What Is Steerability?
Steerability is the ability to influence an AI model's responses through instructions and context.
Instead of receiving the same type of response every time, you can guide how AI explains information, organizes content, or approaches a task.
Steerability makes Generative AI adaptable to different users and purposes.
Steerability means you can guide how AI responds by providing clear instructions about what you want.
2. An Everyday Example: Photosynthesis
Imagine asking an AI assistant:
"Explain photosynthesis."
The AI may provide a general scientific explanation.
Now change the instruction:
"Explain photosynthesis to a 10-year-old student using simple language and one everyday example."
The AI can adapt the explanation to suit the requested audience.
"Plants use sunlight, water, and air to make their own food. Think of sunlight as the energy that helps a plant's tiny food-making system work."
The topic is unchanged. What changes is the way the model presents the explanation.
That is steerability in action.
3. What Aspects of AI Responses Can We Guide?
Instructions can influence many aspects of a generated response.
Audience
Explain a topic for a child, beginner, student, or professional.
Tone
Use friendly, formal, conversational, or professional language.
Format
Organize a response as paragraphs, a table, or steps.
Length
Request a brief summary or a detailed explanation.
Purpose
Explain whether the response is for learning, planning, writing, or decision support.
Providing these details helps the model interpret your expectations.
4. How Do Instructions Influence Responses?
A language model generates text using learned patterns and the context available during the current interaction.
Instructions become part of that context.
When the model generates its next tokens, those instructions can influence the kinds of responses it produces.
How Steerability Works
Explain for a 10-year-old. Use simple language. Include one example.
↓
Uses instructions and context while generating tokens.
↓
A simpler explanation with an everyday example.
A simplified illustration of instruction-guided generation.
Many modern AI assistants also undergo additional training and refinement to improve instruction following.
5. Why Does the Same Question Produce Different Answers?
Consider the question:
"What is climate change?"
Different instructions can lead to different useful responses.
Explain climate change using simple words and a familiar example.
Explain the greenhouse effect and include key scientific concepts.
Summarize climate-related risks and practical response strategies.
The underlying topic remains the same, but the intended audience and purpose change.
Steerability enables an AI model to accommodate these different needs.
6. Does AI Always Follow Instructions?
No.
AI assistants can misunderstand unclear requests, overlook details, or produce unexpected responses.
They may also generate incorrect information while following the requested style or format.
In addition, AI systems may decline instructions that conflict with their safety requirements or higher-priority instructions.
A response can follow your instructions perfectly in tone and format, yet still contain factual errors.
Steerability improves adaptability, but it does not guarantee accuracy, reliability, or successful task completion.
7. How Can We Use Steerability Effectively?
Understanding steerability helps us communicate more purposefully with AI.
A useful approach is to describe the task and provide the essential context.
"Explain photosynthesis to a 10-year-old student in 100 words. Use simple language and include one example from everyday life."
This instruction specifies:
- The topic to explain.
- The intended audience.
- The desired length.
- The language complexity.
- The need for an example.
The model has a clearer description of what a useful response should look like.
After receiving the answer, review it and provide additional feedback if necessary.
Key Takeaway
Steerability allows people to guide AI-generated responses through instructions, context, and feedback.
It makes AI adaptable to different purposes, but instruction following is not always perfect.
Think About It
Imagine explaining the same scientific concept to three different audiences:
- A 10-year-old child.
- A university student.
- A working professional.
How would you change your AI instructions for each audience?
What would remain unchanged in the factual explanation?
What Have You Learned?
After completing Lesson 012, you should be able to:
- Explain the meaning of steerability.
- Describe how instructions influence AI responses.
- Recognize why the same topic can be explained differently.
- Identify ways to guide audience, tone, format, length, and purpose.
- Explain the limitations of AI instruction following.
Module 3 — The Four Properties of AI
Congratulations on reaching the fourth lesson of Module 3!
Together, the four lessons introduce important properties of modern language models.
Lesson 009: Next-Token Prediction — How AI Generates Answers
Lesson 010: Knowledge — What Does AI Actually Know?
Lesson 011: Working Memory — What Is the Context Window?
Lesson 012: Steerability — Why Does AI Follow Instructions?
These properties help explain how AI generates language, uses learned information, processes context, and responds to instructions.
CONTINUE LEARNING
Next: Module 4 — The AI Landscape
Our next module explores different types of AI systems and the roles they play.
You will learn about multimodal AI, chatbots, assistants, AI agents, and the differences between AI, search engines, and databases.
Lesson 013: What Is Multimodal AI?
Further Learning Resources
Your AI Fluency Journey
AI Fluency develops through four connected phases.
AI Fluency begins with understanding AI and progresses toward accomplishing meaningful things with AI.
No comments:
Post a Comment