Learn prompt engineering

{{whyLabel}}: Understanding how AI processes text as tokens rather than words is crucial for managing context windows and costs.

{{howLabel}}:

• Use the OpenAI Tokenizer tool to see how sentences are split into chunks.
• Learn that 1,000 tokens roughly equal 750 words.
• Understand that LLMs predict the next token based on statistical probability, not true 'understanding'.

{{doneWhenLabel}}: You can explain the difference between a word and a token and how it affects context limits.

{{whyLabel}}: These settings control the randomness and creativity of the AI's output.

{{howLabel}}:

• Set Temperature to 0 for factual, consistent tasks (e.g., coding, data extraction).
• Set Temperature to 0.7–1.0 for creative writing or brainstorming.
• Learn that Top-P (nucleus sampling) limits the AI to a percentage of the most likely tokens.

{{doneWhenLabel}}: You can choose the correct parameter settings for a factual vs. a creative prompt.

{{whyLabel}}: Standard chat interfaces (like ChatGPT) hide system settings; Playgrounds offer full control over the model.

{{howLabel}}:

• Create accounts for OpenAI Platform and Anthropic Console.
• Familiarize yourself with the 'System Message' field vs. the 'User Message'.
• Explore the 'Compare' mode to see how different models (e.g., GPT-4o vs. Claude 3.5 Sonnet) handle the same prompt.

{{doneWhenLabel}}: You have access to at least two professional developer consoles.

{{whyLabel}}: Vague prompts lead to vague results; a structured framework ensures the AI has all necessary components.

{{howLabel}}:

• Role: 'Act as a Senior SEO Specialist.'
• Task: 'Write a meta description.'
• Context: 'For a vegan bakery in Berlin.'
• Format: 'Maximum 160 characters, including a call to action.'

{{doneWhenLabel}}: You have rewritten three vague prompts into the RTCF format.

{{whyLabel}}: Providing examples (shots) is the most effective way to teach the AI a specific style or logic without fine-tuning.

{{howLabel}}:

• Provide 3–5 examples of [Input] -> [Output] before your actual request.
• Ensure examples are diverse to prevent the AI from over-fitting to one specific pattern.
• Use 'Zero-Shot' only for very simple, common tasks.

{{doneWhenLabel}}: You have successfully guided an AI to follow a complex custom formatting style using 3 examples.

{{whyLabel}}: Delimiters help the AI distinguish between instructions and the data it needs to process.

{{howLabel}}:

• Use triple quotes ("""), XML tags (<text></text>), or dashes (---) to wrap input text.
• Example: 'Summarize the text found within the <article> tags.'
• This prevents 'prompt injection' where the input text is mistaken for instructions.

{{doneWhenLabel}}: Your prompts clearly separate instructions from data using XML-style tags.

{{whyLabel}}: Telling the AI what not to do is as important as telling it what to do.

{{howLabel}}:

• Use phrases like 'Do not use jargon,' 'Avoid mentioning [Competitor],' or 'Do not apologize for being an AI.'
• Combine this with 'Positive Reinforcement' (e.g., 'Only provide the final answer').

{{doneWhenLabel}}: You have a prompt that generates a response without using five specific 'forbidden' words.

{{whyLabel}}: Forcing the AI to 'think' step-by-step significantly reduces logical errors in math and reasoning.

{{howLabel}}:

• Add the phrase 'Let's think step-by-step' to your prompt.
• For better results, provide a 'Few-Shot CoT' example where you show the reasoning process yourself.
• Use this for debugging code or solving multi-step business problems.

{{doneWhenLabel}}: The AI provides a detailed reasoning chain before giving the final answer.

{{whyLabel}}: For career-level work, AI outputs must often be machine-readable to be integrated into apps.

{{howLabel}}:

• Explicitly ask for 'JSON format'.
• Provide a JSON schema or a template (e.g., '{ "title": "", "summary": "" }').
• Use the 'JSON Mode' setting in the OpenAI API if available.

{{doneWhenLabel}}: You can consistently extract data from a text into a valid, error-free JSON object.

{{whyLabel}}: ToT allows the AI to explore multiple reasoning paths, evaluate them, and backtrack if a path fails.

{{howLabel}}:

• Instruct the AI to: 1. Generate 3 possible solutions. 2. Evaluate the pros/cons of each. 3. Select the best one and expand on it.
• This mimics human brainstorming and leads to more robust strategic plans.

{{doneWhenLabel}}: You have a strategic plan that was selected from three AI-generated alternatives within a single prompt.

{{whyLabel}}: Asking the AI to first identify the high-level principles of a problem helps it solve specific details more accurately.

{{howLabel}}:

• Prompt: 'Before solving [Task], identify the underlying physics/logic principles involved.'
• Then: 'Now, using those principles, solve [Task].'

{{doneWhenLabel}}: You have solved a complex technical question by first generating a 'step-back' abstraction.

{{whyLabel}}: A public portfolio is the #1 way to prove your skills to employers in 2025/2026.

{{howLabel}}:

• Create a repository named 'Prompt-Engineering-Portfolio'.
• Document 'Before' and 'After' prompts with their respective outputs.
• Explain the technique used (e.g., 'Used Few-Shot + JSON Mode for data extraction').

{{doneWhenLabel}}: You have a GitHub repository with at least 5 documented, high-quality prompt use-cases.

{{whyLabel}}: Professional prompt engineering requires objective testing, not just 'vibes'.

{{howLabel}}:

• Create a 'Judge Prompt' that evaluates another AI's output based on specific criteria (accuracy, tone, conciseness).
• Use a scale of 1–10.
• This allows you to automate the testing of hundreds of prompt variations.

{{doneWhenLabel}}: You have a system where one AI model critiques and scores the output of another model.

{{whyLabel}}: Pure 'Prompt Engineer' titles are merging into 'AI Engineer' or 'Workflow Designer' roles.

{{howLabel}}:

• Update your LinkedIn to highlight 'AI Interaction Design' and 'LLM Optimization'.
• Focus on how your prompts save company time or money (e.g., 'Reduced hallucination rate by 40% using CoT').
• Learn basic Python to connect prompts to APIs (using libraries like LangChain or LiteLLM).

{{doneWhenLabel}}: Your resume/LinkedIn reflects prompt engineering as a measurable business optimization skill.