Complete AI Mastery Course

{"contentFormat":"slides.v1","completion":{"requireAllSlides":true,"requireQuiz":true},"slides":[{"kind":"title","eyebrow":"Module 2: Practical Prompt Engineering","title":"Chain-of-Thought Prompting for Complex Tasks","body":"Learn how to guide AI models to reason step by step, improving accuracy on math, logic, planning, and multi-factor decisions. This lesson covers when to use chain-of-thought, how to prompt effectively, and how 2026 reasoning models change the game.","outcomes":["Identify tasks that benefit from chain-of-thought prompting","Write prompts that elicit step-by-step reasoning","Separate reasoning from final answer for clarity","Adapt technique for reasoning models vs. standard models"],"narration":"Welcome to this lesson on chain-of-thought prompting. By the end, you'll be able to guide AI models through complex reasoning tasks and get more accurate results."},{"kind":"content","heading":"What Is Chain-of-Thought Prompting?","body":"Chain-of-thought (CoT) prompting means instructing the model to break a problem into intermediate steps and show its working, rather than jumping straight to a final answer. This technique is one of the most reliable ways to improve performance on tasks that require multiple logical steps.\n\nAnalogy: Ask someone \"What's 17 × 24?\" and demand an instant answer — they may guess wrong. Give them a moment to work it out: 17 × 24 = 17 × 20 + 17 × 4 = 340 + 68 = 408 — and they get it right. A language model behaves similarly: giving it \"room to work\" produces better answers. Each step it writes becomes context that informs the next step.","callout":{"variant":"insight","title":"Why It Works","text":"By generating intermediate steps, the model creates its own scratchpad, reducing errors from skipping steps. This is especially effective for arithmetic, logic, and planning."},"narration":"Chain-of-thought prompting asks the model to reason step by step, like working out a math problem on paper. This reduces errors and makes the reasoning visible."},{"kind":"content","heading":"When to Use Chain-of-Thought","body":"Chain-of-thought helps most on tasks that require multiple reasoning steps. Use it for:\n\n- Arithmetic and word problems — anything with multiple calculations.\n- Logic and deduction — puzzles, conditions, \"if this then that\" chains.\n- Multi-step planning — itineraries, project sequencing, resource allocation.\n- Decisions with trade-offs — comparing options against several criteria.\n\nFor simple lookups (\"What is the capital of Ghana?\") or short creative tasks, CoT adds no value and just makes answers longer.","callout":{"variant":"tip","title":"Quick Check","text":"Ask yourself: Does this task require more than one logical step? If yes, use CoT. If it's a simple fact or single instruction, skip it."},"narration":"Chain-of-thought is powerful for multi-step tasks like math, logic, planning, and decisions. For simple questions, it's unnecessary."},{"kind":"content","heading":"How to Trigger Chain-of-Thought","body":"There are several ways to prompt step-by-step reasoning:\n\n- The classic instruction: Add a line like \"Let's think step by step\" or \"Work through this one step at a time before giving the final answer.\" This alone often improves accuracy.\n- Structure the reasoning: Ask for numbered steps, then a clearly labelled conclusion: \"Show your reasoning as numbered steps, then write 'Final answer:' on a new line.\"\n- Few-shot CoT: For tricky, repeatable problems, include one or two worked examples that show the reasoning, not just the answer. The model imitates the reasoning style.\n\nSeparate the thinking from the final answer: A practical habit is to ask the model to reason first and then give a clean, isolated answer. Example prompt:\n\n

\nSolve the problem. First show your steps. Then, on a new line, write only:\nFinal answer: <value>\n

","callout":{"variant":"exercise","title":"Try It Yourself","text":"Use any chatbot. First ask a multi-step problem without CoT, then with \"Let's think step by step.\" Compare the answers."},"narration":"You can trigger chain-of-thought with simple phrases like 'Let's think step by step' or by structuring the output. Always separate reasoning from the final answer for clarity."},{"kind":"content","heading":"Reasoning Models in 2026","body":"The landscape has shifted. Providers now offer dedicated reasoning models — such as OpenAI's o1 and later reasoning-focused releases, Anthropic's Claude models with an extended-thinking mode, and Google's Gemini \"thinking\" variants. These models perform an internal chain-of-thought automatically before responding, so you often don't need to write \"think step by step\" yourself.\n\n- With a reasoning model: state the problem clearly and let it reason internally. You can still ask it to show its steps if you want to verify them.\n- With a standard/fast model (e.g. GPT-4o, a smaller open Llama model): explicit CoT instructions give the biggest boost.\n\nImportant caution: showing steps makes reasoning visible, but visible reasoning is not proof of correctness. A model can produce a confident, well-formatted chain that still reaches a wrong answer. Always sanity-check the conclusion, especially for numbers, money, dates, and legal or medical matters.","callout":{"variant":"warning","title":"Sanity Check Required","text":"Visible reasoning is not proof of correctness. Always verify final answers, especially for critical applications."},"narration":"In 2026, reasoning models like o1 and Claude extended-thinking handle chain-of-thought internally. For standard models, you still need explicit prompts. But always verify the output."},{"kind":"content","heading":"Real-World Examples","body":"Chain-of-thought prompting is used globally across industries:\n\n1. Logistics planner, Lagos (Nigeria): Tunde must sequence deliveries across five districts within fuel and time limits. A CoT prompt — \"List the constraints, evaluate each route step by step, then recommend an order\" — surfaces the reasoning so he can verify it respects the fuel budget.\n\n2. Accounting student, Ho Chi Minh City (Vietnam): Linh practises multi-step VAT and margin calculations. With \"Solve step by step, showing each calculation,\" she sees exactly where a figure goes wrong and learns from it.\n\n3. NGO grant reviewer, Amman (Jordan): A reviewer scores proposals against four weighted criteria. A CoT prompt asks the model to assess each criterion in turn, assign a score with a one-line justification, then total them — turning a vague \"this looks good\" into a transparent, auditable evaluation.","cards":[{"title":"Logistics (Nigeria)","text":"CoT helps sequence deliveries under constraints."},{"title":"Accounting (Vietnam)","text":"Step-by-step calculations reveal errors for learning."},{"title":"Grant Review (Jordan)","text":"Transparent scoring against weighted criteria."}],"narration":"These examples from Nigeria, Vietnam, and Jordan show how chain-of-thought helps in logistics, education, and decision-making."},{"kind":"content","heading":"Common Pitfalls & How to Avoid Them","body":"Avoid these mistakes to get the most out of chain-of-thought:\n\n| Pitfall | Fix |\n|---------|-----|\n| Using CoT everywhere | Reserve for genuinely multi-step tasks |\n| Trusting visible steps as proof | Independently verify final numbers, dates, and facts |\n| Mixing reasoning and answer together | Ask for a clearly labelled \"Final answer:\" line at the end |\n| Adding \"think step by step\" to a reasoning model unnecessarily | For reasoning models, just state the problem; let them reason internally |","table":{"headers":["Pitfall","Fix"],"rows":[["Using CoT everywhere","Reserve for genuinely multi-step tasks"],["Trusting visible steps as proof","Independently verify final numbers, dates, and facts"],["Mixing reasoning and answer together","Ask for a clearly labelled \"Final answer:\" line at the end"],["Adding \"think step by step\" to a reasoning model unnecessarily","For reasoning models, just state the problem; let them reason internally"]]},"narration":"Watch out for these common pitfalls: overusing CoT, trusting reasoning blindly, and mixing output. Use the table as a quick reference."},{"kind":"quiz","heading":"Check Your Understanding","questions":[{"question":"When should you use chain-of-thought prompting?","options":["For all prompts to ensure thoroughness","Only for tasks that require multiple logical steps","Only when using reasoning models","Never, as it makes answers too long"],"questionId":"cmrf73k49000lpd27pgr5vvhn"},{"question":"Which of the following is a correct way to prompt chain-of-thought?","options":["Add 'Be concise' to the prompt","Include 'Let's think step by step' before the question","Ask for the answer only, then request reasoning separately","Use a reasoning model and never ask for steps"],"questionId":"cmrf73k49000mpd27ea2vuvdn"},{"question":"True or False: Visible reasoning from a model is always correct.","options":["True","False"],"questionId":"cmrf73k49000npd27lwsxd4oo"}],"narration":"Let's test your understanding with a quick quiz.","quizId":"qz_cmk7lfzd80007g4p8v2y3380v"},{"kind":"summary","heading":"Key Takeaways","takeaways":["Chain-of-thought prompting improves accuracy on multi-step tasks by having the model reason step by step.","Use explicit instructions like 'Let's think step by step' or structure the output with numbered steps and a final answer line.","Reserve CoT for tasks requiring reasoning; skip it for simple lookups or creative tasks.","In 2026, reasoning models (o1, Claude extended-thinking, Gemini thinking) handle CoT internally, but you can still ask for visible steps.","Always verify the final answer independently — visible reasoning is not proof of correctness."],"narration":"In summary, chain-of-thought prompting is a powerful technique for complex tasks. Use it wisely, verify outputs, and adapt to the model you're using."}]}