AI · Beginner
AI & Prompting 101
How LLMs actually work and how to write prompts that hold up.
What you will learn
The 10 topics inside AI & Prompting 101
Each topic has its own goal, its own questions and its own mastery score, so you always know which part is still weak.
Failure modes
Recognise hallucination, drift and prompt injection in the wild.
Iterative prompt refinement
Test and revise prompts systematically instead of by feel.
Multimodal prompting
Design prompts that combine text with images and other inputs.
Prompt anatomy
Name the parts of a prompt and what each part is for.
Responsible use
Apply review, disclosure and data-handling habits.
Sampling parameters
Predict how temperature and top-p change the output.
Step-by-step reasoning
Know when asking for reasoning helps and when it wastes tokens.
Tokens, context and prediction
Explain what the model is actually doing when it answers.
Working with model APIs
Explain roles, messages and the shape of an API request.
Zero-shot vs few-shot
Choose between describing the task and demonstrating it.
Example questions
Real questions from this course
These are taken straight from the course. Pick an answer, then open the explanation to see why it is right.
Sample 01
What is the primary role of systematic error analysis when analyzing failures in an evaluation run during prompt iteration?
Show the answer and why
Answer: Performing error categorization identifies recurring failure patterns to guide targeted and evidence-based prompt modifications.
Defends against making arbitrary prompt modifications by feel, demonstrating that systematic error analysis informs structured, targeted prompt iterations.
Systematic error categorization involves grouping failures into specific classes, such as missing instructions, formatting errors, or hallucinated details. Identifying these pattern clusters guides targeted prompt revisions rather than making random, ungrounded changes.
Sample 02
What security risk occurs when a language model processes untrusted third-party data containing hidden malicious instructions?
Show the answer and why
Answer: Susceptibility to indirect prompt injection through embedded malicious instructions in retrieved text.
Indirect prompt injection occurs when untrusted data sources manipulate model instructions without direct user input. This counters the misconception that prompt injection only matters for public chatbots.
Processing untrusted external data that contains malicious directives exposes the system to indirect prompt injection. Attackers embed instructions in documents or web pages to hijack the model's behavior.
Sample 03
Why is prompt engineering for image inputs different from purely text-based prompting?
Show the answer and why
Answer: Multimodal prompt structure often requires placing images before text questions, reflecting how visual token grids are processed differently from text.
The answer explains that vision inputs require deliberate spatial ordering and formatting within payloads. It defends against the misconception that image prompts behave exactly like text prompts, which causes users to treat image payloads identically to text strings.
Effective multimodal prompt structure requires placing visual inputs before text instructions or explicitly pointing to visual coordinates, as images are processed as token grids rather than sequential prose. Assuming image inputs behave identically to text inputs leads to poor contextual alignment and ambiguous queries.
AI & Prompting 101 is free with an account, along with every other course in the library.
Start free