Understanding AI: Foundations for Strategic Application
A practical guide by Kelly "Curly" Ihme on implementing AI tools to enhance leader workflows, research, and training effectiveness.
*Created in collaboration with artificial intelligence tools
QR for this site
What We're Focusing On
What We ARE Covering
  • Practical, useful tools (free/cheap/safe)
  • AI compatibility with systems
  • Data security considerations
  • Quality resources for further learning
What We're NOT Covering
  • Detailed explanations of AI types
  • AI boom, arms race, TechBros
  • "Cyber"
  • MATH
Understanding AI and the Rise of Generative AI
4
Artificial Intelligence (AI)
AI is a technology that enables machines to mimic human intelligence, with 4 general categories:
numerical prediction, classification, robotic navigation, and language processing.
Generative AI
A subset of AI that creates new content like text, images, music, or code by learning patterns from existing data to generate realistic outputs.
Key Differences from Traditional AI:
• Traditional AI: Analyzes or classifies data
• Generative AI: Generates new content using learned patterns
Gen AI Common Models:
  • GANs (Generative Adversarial Networks): Create realistic images
  • GPTs (Generative Pre-trained Transformers): Generate human-like text)
  • Agentic AI (AI that can take actions and complete multi-step tasks on its own, not just answer questions).
4
Basic Prompts
Custom Instructions
RAG: Retrieval Augmentation Generation
Custom GPTs (Gems, Agents)
RLM: Recursive LM
How Large Language Models Actually Work
Understanding the mathematics behind AI-powered language prediction
The Core Mechanism

LLMs don't 'understand' language—they predict the most statistically probable next word based on patterns learned from billions of text examples.

Modern LLMs like GPT-4 have over 1 trillion parameters—each one a mathematical weight that helps predict language patterns with remarkable accuracy.
Training Phase
Models analyze massive datasets (books, articles, websites, AI conversations) to learn statistical patterns in language syntax and word relationships
Mathematical Encoding
Text is converted into numerical vectors in high-dimensional space where similar concepts cluster together
Probability Calculation
Neural networks with billions of parameters compute probability distributions for what word should come next
Output Generation
The model selects words based on these probabilities, creating human-like text through pure mathematical prediction
Tokens & Vectors
Words broken into pieces and represented as numbers
Attention Mechanisms
Mathematical functions that weigh which previous words matter most
Parameters
Billions of adjustable weights fine-tuned during training
Understanding Context Windows: Why AI Conversations Degrade
The memory limitations that affect long AI interactions
What is a Context Window?

Context windows are the 'working memory' of an LLM—a fixed limit on how much text the model can process at once, measured in tokens.

Pro Tip: For complex tasks, start fresh conversations periodically or provide key context in each message to maintain quality as conversations lengthen.
Fresh Conversation Start
Model has full context window available. All instructions and details are 'remembered' and influence responses.
Window Fills Up
As conversation grows, older messages get pushed out. The model can only 'see' the most recent exchanges within its token limit.
Context Loss Begins
Early instructions, key details, and nuanced context disappear from the model's view, even though you still see them.
Performance Degradation
Without access to earlier context, responses become less accurate, repetitive, or miss important constraints you established.
Forgotten Instructions
Initial guidelines and preferences established early in conversation become invisible to the model
Lost Continuity
References to earlier topics or decisions may be missed or contradicted
Reduced Coherence
Long conversations may feel disjointed as the model loses the narrative thread
AI as a Decision-Support Tool đź’»

Generating 🚫, Accelerating ⏩, Communicating 💬

File upload

AY26 USAWC Generative AI Policy.pdf

223.8 KB

"By far, the greatest danger of Artificial Intelligence is that people conclude too early that they understand it." -Eliezer Yudkowsky, AI Researcher and Writer
Challenges in Working with AI
AI should support human decision making and judgment…not replace it!
Hallucinations & Inaccuracy
AI can generate false information confidently because of context window limitations and the predictive math that underlies LLMs.
Data Privacy & Security
Sensitive information can be exposed through prompts, risking operational security. Sovereign AI or closed models can mitigate risk.
Over-Reliance & Skill Degradation
Depending too heavily on AI can erode critical thinking and core competencies.
Bias & Ethical Concerns
AI models can perpetuate biases and raise ethical questions about appropriate use.
Gen AI: Relevance and Core Capabilities
Military Applications
  • Drafting: Quickly generate operational plans, OPORD fragments, or intel summaries for staff review
  • Scenario Generation: Prototype COAs, simulate adversary reactions, or generate wargame injects
  • Planning Support: Outline logistics concepts, terrain analyses, and pre-brief talking points
Generative AI extends traditional AI capabilities to new paradigms
  • Multi-modal input and output
  • Rapid content generation
  • Dynamic modeling and simulation
  • Creative ideation and innovation
Academic Applications
  • Research: Summarize doctrine, journal articles, and historical case studies
  • Writing: Draft outlines or discussion posts; generate thesis statements or counterarguments
AI Basics #1
Managing Your AI Workflows
Rename AI conversations to your topic area for better organization; pin key conversations
Maintain Context
Return to the same conversation as you work on a project, update your context window
Reference Previous Discussions
Use prompts like "Using the previous discussion..." for continuity
Model Alignment
Each AI has specialty functionality, keep like functions with the same AI (images with Dall-E, code with Claude, videos with Veo and Nano Banana, research with Thesify, etc)
AI Basics #2
Prompt Design
The Answer is 42
Role/Persona
Assign the AI a specific role like "expert historian" or "UX designer" to shape response style.
Specificity
Include precise details about format, length, and tone. Vague prompts yield vague results.
Example Output
Show the AI what success looks like with sample formats or structures.
Context
Provide relevant background information to help the AI understand your needs.
Feedback
Refine results by telling the AI what worked and what needs improvement.

Bad Prompt: “Write a plan for training”

Good Prompt: "You are a military training officer assigned to develop a two-day urban warfare training exercise for Task Force Alpha in the fictional city of Fortwood. The primary objective is to train on building clearing, coordination between infantry and armored units, and communication with air support. Please provide a structured plan, including a timeline, key tasks, and expected outcomes for each phase. You may use examples of past urban training operations for context. If you need more details about the unit’s composition, available resources, or exercise limitations, ask clarifying questions before proceeding."

AI Basics #3
Efficient AI-to-Document Workflow
GenAI.mil output
Use a GPT or other AI tools to create initial content
Copy AI output and paste into Excel for table manipulation or Word for documents
Embed the formatted content into your articles, outlines, or other materials
AI Basics #4:
Converting Handwritten Notes to Text
Take a clear photo of your handwritten notes or sketches
Upload to a GPT
Create memos, notes, images
Capture brainstorming sessions from whiteboards or multiple notetakers
Create summaries, MFRs, and reports
AI Basics #5
Research and Source Generation
Specialized AI Tools
Use purpose-built research tools like Elicit AI, Thesify, and AI Agenics to find sources and generate literature reviews with proper citations.
2
Craft detailed prompts for general LLMs or GPTs to develop comprehensive source lists and research summaries.
3
Verification Process
Always verify AI-generated sources and cross-reference information for accuracy before including in any work.
4
Use Specialized Models
Deep Research models without GenAI.mil (CAC required) and commercial platforms provide the "thinking" and logic steps along with citations, links, and other sources as prompted for detailed answers.
Advanced AI Incorporation
Enhancing Engagement
Create realistic conversation scenarios for student practice
Summarize Content
Condense complex discussions into key takeaways
Address Learning Styles
Adapt content to different learner modalities
Promote Critical Thinking
Develop thought-provoking questions and scenarios
Socratic Tutor
Streamlining Training Development
Lesson Planning
Generate structured lesson outlines and objectives
Develop engaging materials and activities
Create varied assessment tools and rubrics
Refinement
Improve based on outcomes and feedback

ChatGPT

ChatGPT - 6-Week Peak Mind Syllabus

Shared via ChatGPT

The Future of AI
Exponential Growth
AI capabilities continue to follow Moore's Law, with processing power doubling approximately every two years.
Creative AI
Video and song creation tools now produce professional-quality content with minimal human input.
Intelligent Gaming
Games feature increasingly sophisticated AI opponents and companions that adapt to player behavior.
Human-Machine Teaming
The future workforce will blend human creativity with AI efficiency in seamless partnerships.
Questions?
Ethan Mollick on LinkedIn & Substack
Follow for updated opinions on which AI tools to use and practical applications in education. Visit: https://www.oneusefulthing.org/p/which-ai-to-use-now-an-updated-opinionated
Andy Stapleton on YouTube
Excellent tutorials and demonstrations of AI tools in action. Watch: https://www.youtube.com/watch?v=QDZDPnTYOZg&t=624s&pp=ygUKbm90ZWJvb2tsbQ%3D%3D
How AI Works on YouTube
8 minute video explaining large language models. Watch: https://www.youtube.com/watch?v=LPZh9BOjkQs
19 minute video on neural networks. Watch: https://www.youtube.com/watch?v=aircAruvnKk&t=2s
Additional Learning Materials
Explore books, courses, and other resources to deepen your understanding of AI applications in education.
AI in PME article by Ihme and Rasmussen
AI in OSINT article by Ihme