Day 2 - Session 1: Basic Prompting Foundations

How do we populate the context window effectively?

Author

Dr Brian Ballsun-Stanton

Published

September 9, 2025

Good Morning!

Today’s Big Question

How do we populate the context window effectively?

This Morning

Share your annotated prompts
Learn basic prompting theory
Practice iterative questioning
Begin systematic annotation

Show Me Your Prompts! (15 min)

What We’ll Learn This Session

By the end of this morning, you will be able to:

Explain: How iterative questioning builds context
Identify: Which AI questions advance vs. derail your work
Track: When prompts succeed vs. fail through annotation

Basic Theory: Rules of Thumb

The Engineering Method

“Solving problems using rules of thumb that cause the best change in a poorly understood situation using available resources” - Bill Hammack

For Prompting, This Means

We don’t know exactly why things work
We develop local heuristics through practice
What works depends on context
Iteration beats perfection

Key Prompting Principles

1. Context is Everything

The AI only knows what’s in the current conversation

2. Specificity Matters

Vague instructions → vague outputs

3. Iteration is Expected

First attempts rarely perfect

4. Structure Helps

Break complex tasks into steps

Exercise: Weakening the Prompt (15 min)

Remember Yesterday’s Prompt?

Take the “ask me one question at a time” prompt from yesterday.

Now Break It

Try: “Help me figure out my goals for the week” without any of the setup.

Compare Results

What changes? Put up green sticky when complete.

What Makes Prompts Effective?

Good prompts

One task at a time
High scaffolding
- Is it clear what the intention of the conversation is?
- Is it clear what the register is?
- Are the answers easily falsifable?

Bad Prompts

Generic
Fact based
Poorly populated context window
Multiple questions

Annotation Practice (20 min)

On the Conceptboard

With your conversation from the weakening exercise:

Your prompts: Mark what instructed useful behavior
AI responses: Note where it followed/ignored instructions
Patterns: What words consistently trigger better responses?

The Context Window

Think of it as a bucket: - Everything must fit inside - New information pushes out old - Quality matters more than quantity - You control what goes in

Your prompts are the recipe for filling this bucket effectively.

Exercise: Playing with GPT-2 (10 minutes)

Go visit mirror.zad-giessen.de/perplexity

We will talk about why the context window matters so much as you play with it.

Other concepts:

Tokens
Temperature

Thou shalt not allow an error to live.

Looking Ahead

Today: Metaprompting

Can AI write its own prompts?
The blank page problem
Epistemic humility (or lack thereof)

Tomorrow: What Can We Verify?

Working with documents
Extracting vs. interpreting
Model differences matter

Key Takeaways

Prompting is engineering - rules of thumb, not laws
Iteration is required - expect to refine
Context accumulates - each exchange builds on the last. Do not allow errors to remain.
Your judgment develops - through annotation and reflection

Before the Break

Save your annotated conversations
Add all prompts to our grimoire
- Let us know which ones worked and which ones didn’t
Think about: What patterns are you noticing?

See you at 11:00 for metaprompting!

--- title: "Day 2 - Session 1: Basic Prompting Foundations" subtitle: "How do we populate the context window effectively?" author: - "Dr Brian Ballsun-Stanton" date: "9 September 2025" format: revealjs: slide-number: true chalkboard: true theme: [default, ../reveal-custom.scss] footer: "AI Summercamp 2025 - Basic Course" preview-links: auto transition: slide background-transition: fade highlight-style: github navigation-mode: vertical --- ## Good Morning! ### Today's Big Question **How do we populate the context window effectively?** ### This Morning - Share your annotated prompts - Learn basic prompting theory - Practice iterative questioning - Begin systematic annotation --- ## Show Me Your Prompts! (15 min) ### Our First Sharing Session 1. Who tried the homework prompt? 2. What surprised you about the AI's questions? 3. What did you highlight in pink (didn't work)? 4. What did you highlight in blue (worked well)? 5. What other prompts did you experiment with overnight? **Remember**: We learn from failures as much as successes --- ## What We'll Learn This Session By the end of this morning, you will be able to: - **Explain**: How iterative questioning builds context - **Identify**: Which AI questions advance vs. derail your work - **Track**: When prompts succeed vs. fail through annotation --- ## Basic Theory: Rules of Thumb ### The Engineering Method "Solving problems using rules of thumb that cause the best change in a poorly understood situation using available resources" - Bill Hammack ### For Prompting, This Means - We don't know exactly why things work - We develop local heuristics through practice - What works depends on context - Iteration beats perfection --- ## Key Prompting Principles ### 1. Context is Everything The AI only knows what's in the current conversation ### 2. Specificity Matters Vague instructions → vague outputs ### 3. Iteration is Expected First attempts rarely perfect ### 4. Structure Helps Break complex tasks into steps --- ## Exercise: Weakening the Prompt (15 min) ### Remember Yesterday's Prompt? Take the "ask me one question at a time" prompt from yesterday. ### Now Break It Try: "Help me figure out my goals for the week" without any of the setup. ### Compare Results What changes? Put up green sticky when complete. --- ## What Makes Prompts Effective? ### Good prompts - One task at a time - High scaffolding - Is it clear what the intention of the conversation is? - Is it clear what the register is? - Are the answers easily falsifable? ### Bad Prompts - Generic - Fact based - Poorly populated context window - Multiple questions --- ## Annotation Practice (20 min) ### On the Conceptboard With your conversation from the weakening exercise: 1. **Your prompts**: Mark what instructed useful behavior 2. **AI responses**: Note where it followed/ignored instructions 3. **Patterns**: What words consistently trigger better responses? ### Share with Your Neighbor Compare annotations - same patterns? --- ## The Context Window Think of it as a bucket: - Everything must fit inside - New information pushes out old - Quality matters more than quantity - You control what goes in Your prompts are the recipe for filling this bucket effectively. --- ## Exercise: Playing with GPT-2 (10 minutes) Go visit [mirror.zad-giessen.de/perplexity](https://mirror.zad-giessen.de/perplexity) We will talk about why the context window matters so much as you play with it. Other concepts: - Tokens - Temperature **Thou shalt not allow an error to live.** --- ## Looking Ahead ### Today: Metaprompting - Can AI write its own prompts? - The blank page problem - Epistemic humility (or lack thereof) ### Tomorrow: What Can We Verify? - Working with documents - Extracting vs. interpreting - Model differences matter --- ## Key Takeaways 1. **Prompting is engineering** - rules of thumb, not laws 2. **Iteration is required** - expect to refine 3. **Context accumulates** - each exchange builds on the last. Do not allow errors to remain. 4. **Your judgment develops** - through annotation and reflection --- ## Before the Break - Save your annotated conversations - Add all prompts to our grimoire - Let us know which ones worked and which ones didn't - Think about: What patterns are you noticing? See you at 11:00 for metaprompting!