Day 2 Overview: Context & CO-STAR

Duration: 1.5 hours
Focus: Understanding the critical role of context and learning the CO-STAR framework

Learning Objectives

By the end of Day 2, you will:

✅ Understand why context is the foundation of effective prompting
✅ Apply the CO-STAR framework for strategic communication
✅ Identify and verify hallucinations in AI outputs
✅ Navigate the trade-offs between AUTOMAT and CO-STAR
✅ Create strategic documents that match audience needs

Session Structure

Part 1: Context Deep Dive (30 minutes)

Why Context Matters (10 min) - LLMs navigate embedding spaces - Vague vs. precise context - Types of context (explicit, implicit)

Context Best Practices (10 min) - Making implicit context explicit - Iterative context building - Working within context limits

Hands-On Practice (10 min) - Transform vague prompts - Test in sandbox - Compare outputs

Part 2: CO-STAR Framework (40 minutes)

Framework Introduction (15 min) - Six components: Context, Objective, Style, Tone, Audience, Response - When to use CO-STAR vs. AUTOMAT - Strategic communication principles

Materials Science Applications (15 min) - Investor pitch documents - Literature reviews - Strategic reports - Complete CO-STAR examples

Hands-On Practice (10 min) - Build your first CO-STAR prompt - Test in sandbox - Refine based on output

Part 3: Hallucination Detection (20 minutes)

Advanced Verification (10 min) - Beyond basic citation checks - Quantitative data verification - Logical consistency analysis - The "Skeptical Colleague" test

Hallucination Hunt Workshop (10 min) - Identify errors in AI-generated reports - Learn verification protocols - Practice critical evaluation

Key Concepts

1. Context is NOT Optional

Without context: - LLMs explore vast embedding spaces randomly - Generic, off-target responses - Inconsistent outputs

With precise context: - Directed navigation through knowledge space - Relevant, targeted responses - Reproducible quality

2. CO-STAR for Rich Communication

Component	Purpose
Context	Background and situation
Objective	What success looks like
Style	Writing approach
Tone	Emotional quality
Audience	Who will consume this
Response	Format and structure

3. AUTOMAT vs. CO-STAR Decision Matrix

Use AUTOMAT when: - Functional, structured tasks - Clearly defined outputs (tables, code) - Minimal context needed

Use CO-STAR when: - Narrative or strategic content - Rich contextual understanding required - Audience considerations complex

4. Hallucinations Are Inevitable

Accept reality: - LLMs are pattern-matching engines, not databases - They will fabricate plausible-sounding information - Confidence does NOT equal accuracy

Your responsibility: - Verify every citation - Check quantitative claims - Test logical consistency - Apply domain expertise

Today's Exercises

You'll practice with:

Context Transformation
Take vague prompts and add explicit context
A/B test in sandbox
Measure quality improvement
CO-STAR for Strategic Communication
Write memo to R&D Director about AI adoption
Address accuracy, IP security, training concerns
Persuade scientifically-minded skeptics
Hallucination Hunt
Find 5 errors in AI-generated synthesis report
Apply verification protocols
Practice critical evaluation
Framework Selection Challenge
Given 5 scenarios, choose AUTOMAT or CO-STAR
Justify your choice
Build appropriate prompts

What You'll Build Today

Template Library (Expanded)

Add to your collection: - Context checklist for strategic documents - CO-STAR template for investor/management communication - Hallucination verification protocol - Framework selection flowchart

Real Efficiency Gains

Traditional approach: - Strategic document drafting: 4-6 hours - Multiple revision cycles - Inconsistent quality

With CO-STAR: - Strategic document drafting: 1-2 hours (65% reduction) - Fewer revision cycles (clear specification upfront) - Consistent, audience-appropriate quality

Key Distinctions

Context vs. Background

Background: General information about a topic
Context: Specific information relevant to this task

Example:

Background (too broad):

AmaDema is a nanotechnology company founded in 2022...
[500 words of company history]

Context (task-relevant):

For this investor pitch, emphasise our 40% carbon footprint 
reduction vs. PLA and 3 industry partnerships (aerospace, 
automotive, biomedical). Target audience is ESG-focused VCs.

Style vs. Tone

Style: The writing approach (academic, journalistic, technical)
Tone: The emotional quality (confident, cautious, inspirational)

You can mix: - Academic style + Inspirational tone - Technical style + Cautious tone - Journalistic style + Confident tone

Common Questions

"How much context is too much?"

Rule of thumb: Include context that: ✅ Affects the output quality
✅ Clarifies ambiguity
✅ Defines success criteria

Exclude context that: ❌ Is interesting but irrelevant to this task
❌ Would be obvious to your audience
❌ Doesn't change the output

"When should I use CO-STAR vs. AUTOMAT?"

Quick test:

Ask: "Does audience perspective significantly change the output?" - YES → CO-STAR (report for investors ≠ report for engineers) - NO → AUTOMAT (data extraction table is same regardless)

Ask: "Is the output primarily narrative?" - YES → CO-STAR - NO → AUTOMAT

"How do I verify AI outputs efficiently?"

Priority-based verification:

Critical claims (will be challenged by reviewers/stakeholders) → Full verification
Quantitative data (numerical values, citations) → Spot-check verification
General statements (common knowledge) → Minimal verification

Use your domain expertise to focus verification efforts where it matters.

Pre-Work Review

If you completed the pre-work, you've already seen: - What are LLMs? - Context Matters

Today we'll go deeper: - From understanding why context matters to how to build it - From basic context to strategic CO-STAR framework

Looking Ahead

Day 3 will shift focus: - Technical architecture (how LLMs actually work) - NLMs vs. LLMs (why this generation is different) - Green AI introduction (environmental impact)

Day 4 will cover: - Optimization strategies (reduce computational waste) - Advanced conversational learning - Ethics, bias, and responsible deployment

Success Criteria

You're ready for Day 3 when you can:

✅ Explain why vague context leads to poor outputs
✅ Write a complete CO-STAR prompt for a strategic document
✅ Choose appropriately between AUTOMAT and CO-STAR
✅ Identify and verify hallucinations systematically
✅ Build iterative context refinement workflows

Let's Begin!

Ready to master context and strategic communication?

Next: Context Deep Dive →