I completed the AI Fluency : Framework and Foundations from Anthropic. Here are my notes

Intro to AI Fluency

• AI can be a trusted partner for creative and innovative problem-solving.
• Aim to be effective, efficient, ethical, and safe.

• Three distinct ways we interact with AI:

  • Automation: AI does work for humans (e.g., summarising emails).
  • Augmentation: you and AI work together.
  • Agency: AI works independently on your behalf.
• Generative AI creates new content rather than only analysing existing content.
• A simple evaluation lens for LLMs: Helpful, Honest, Harmless.

Weaknesses of Generative AI

• Knowledge cut-off dates.
• Doesn’t verify training data; prone to mistakes.
• Hallucinations (misinformation).
• Limited by context-window size.
• Non-deterministic outputs.

The 4D Fluency Framework

D1 — Delegation: Human vs AI — who does what?

Three aspects:

• Problem awareness

  • Clearly define goals and the work required.
  • Specify what “success” looks like.
  • Identify the kind of thinking and work needed.

• Platform awareness

  • Working knowledge of capabilities and limitations.
  • Choose models that fit the task.
  • Prioritise what matters most: speed, creativity, depth, or accuracy.

• Task delegation

  • What can be usefully automated?
  • Where would augmentation add more value?
  • What should be done by a human alone?
  • What could an agent do on your behalf?

D2 — Description: Communicating with AI

• Don’t just write prompts—explain tasks, ask questions, provide context, and guide the interaction.
• Build a shared thinking environment.

Three aspects:

• Product description

  • Describe the characteristics of the desired output.
  • Be clear about what you want.
  • Include context, format, audience, style, and constraints.
  • Give AI all the information it needs to deliver.

• Process description

  • Guide the AI’s thought process; the “how” can matter more than the “what”.
  • Provide training specific to your problem.
  • Specify data, key tasks, and preferred order.

• Performance description

  • Define behavioural expectations.
  • Note that AI may behave differently by context.
  • State how you want the AI to behave.

D3 — Discernment: Evaluate what AI produces, how it produces it, and how it behaves

• Use domain expertise.
• Understand how AI systems work and their typical shortcomings.

Three aspects:

• Product discernment

  • Factually accurate?
  • Appropriate for the audience and purpose?
  • Coherent and well-structured?
  • Meets requirements?
  • Adds value?

• Process discernment

  • Look for logical inconsistencies.
  • Watch for lapses in attention or inappropriate steps.
  • Note when it gets stuck in small details or circular reasoning.

• Performance discernment

  • Is the communication style appropriate?
  • Is the information at the right level?
  • Does it respond appropriately to feedback?
  • Is the interaction efficient?

Feedback and correction

• Specify the problem.
• Clearly explain what’s wrong.
• Offer concrete suggestions for improvement.
• Revise instructions or examples as needed.

D4 — Diligence: Taking responsibility for your AI interactions

• Be rigorous, transparent, and accountable.
• Consider broader ethical and practical questions.
• Responsibility starts with awareness.

Three aspects:

• Creation diligence: Which AI systems you choose and how you use them.

  • The AI system(s) you use.
  • How you work with them.
  • The impacts from interaction.

• Transparency diligence: Be open and accurate about AI use with stakeholders.

  • Who needs to know.
  • How to communicate it.
  • What level of detail is needed.

• Deployment diligence: Informed responsibility for outputs you use.

  • Verify facts.
  • Check for biases.
  • Ensure accuracy.
  • Confirm usage rights.

Foundational Prompting Techniques

• Provide context: what you want, why you want it, and who you are.

• Offer examples (n-shot prompting): show what “good” looks like; cover a range of cases or styles.

  • Use step-by-step (chain-of-thought) when helpful.
• Specify output constraints: set format, sections, length, and other limits.
• Break complex tasks into steps: use step-by-step where it adds clarity.
• Ask the AI to think first: e.g., “Before answering, think this through carefully.”
• Define role, style, or tone.

Make it iterative

• Ask the AI for help with prompting.
• Effective prompting is iterative: Preliminary prompt → AI response → Refine prompt → Final output.
• Ask for variations.
• Request different formats.
• Check confidence.
• Reset the conversation when needed.