✨ Prompt Engineering Foundations

• What a Prompt Really Is (8 min) — ⚡ A prompt is the full input you give a model: the task, the context, any examples, and the format you want back. The m…
• Roles & System Prompts (8 min) — ⚡ Assigning a role reliably shifts tone, vocabulary, depth, and priorities. "You are a senior copywriter who writes pun…
• Few-Shot: Teaching by Example (8 min) — ⚡ Few-shot prompting means including a handful of input→output examples so the model infers the pattern instead of rely…
• Chain-of-Thought: Reasoning Out Loud (8 min) — ⚡ Chain-of-thought (CoT) prompting asks the model to generate intermediate reasoning steps before its final answer. Bec…
• Controlling the Output Format (8 min) — ⚡ When output feeds a person who needs structure — or worse, another program — you must constrain the format, or the ne…
• Be Specific: Constraints, Criteria & Context (8 min) — ⚡ Most disappointing outputs come from the model filling gaps you left open. The cure is to supply constraints, success…
• Iterating & Debugging Prompts (8 min) — ⚡ Prompting is iterative. Expert prompters don't write perfect prompts; they debug quickly. The skill is reading a bad …
• Pitfalls & Prompt Injection (9 min) — ⚡ As you get fluent, the failures shift from 'I didn't know how' to a handful of recurring pitfalls: • Ambiguity: the t…

What a Prompt Really Is

⚡ A prompt is the full input you give a model: the task, the context, any examples, and the format you want back. The model has broad knowledge but zero awareness of your specific goal until the prompt supplies it. Everything in the prompt steers the output; nothing outside it e…

Roles & System Prompts

⚡ Assigning a role reliably shifts tone, vocabulary, depth, and priorities. "You are a senior copywriter who writes punchy, concrete B2B copy" produces measurably different output than no role at all — it primes the patterns associated with that expertise. There are two channels…

Few-Shot: Teaching by Example

⚡ Few-shot prompting means including a handful of input→output examples so the model infers the pattern instead of relying on your description. Zero-shot is just asking; few-shot shows. It's the fastest way to nail a specific format, tone, or labeling scheme. Example — classifyi…

Chain-of-Thought: Reasoning Out Loud

⚡ Chain-of-thought (CoT) prompting asks the model to generate intermediate reasoning steps before its final answer. Because each token is conditioned on the previous ones, producing the reasoning gives the model 'room' to work, which markedly improves accuracy on multi-step task…

Controlling the Output Format

⚡ When output feeds a person who needs structure — or worse, another program — you must constrain the format, or the next step can't rely on it. Free-form prose is fine for reading; it's a liability for automation. Techniques, weakest to strongest: • Ask plainly: "Respond as a m…

Be Specific: Constraints, Criteria & Context

⚡ Most disappointing outputs come from the model filling gaps you left open. The cure is to supply constraints, success criteria, and context up front. • Constraints bound the output: length, reading level, format, what to include or avoid, tone, time/scope. • Success criteria t…

Iterating & Debugging Prompts

⚡ Prompting is iterative. Expert prompters don't write perfect prompts; they debug quickly. The skill is reading a bad output, diagnosing the specific cause, and changing one thing at a time. A debugging loop: 1. Run the prompt and read the output critically against your success…

Pitfalls & Prompt Injection

⚡ As you get fluent, the failures shift from 'I didn't know how' to a handful of recurring pitfalls: • Ambiguity: the top cause of weak output — fix with specificity (lessons 1 & 6). • Conflicting instructions: 'comprehensive but one sentence', 'formal yet casual'. The model spl…