Why prompt engineering is necessary
How a language model produces text A large language model (LLM) predicts the next token from the tokens before it, one token at a time. It has no lookup of facts; it produces likely text.
Term Meaning Effect on prompting Token A chunk of text, often part of a word; about 4 characters of English on average Limits and costs are counted in tokens Context window Maximum tokens the model can consider at once, prompt plus reply Long documents may be cut off or ignored Training cutoff Date the training data ends Recent events are unknown unless supplied or searched Temperature Setting for randomness in token choice Low: consistent, focused. High: varied, creative Multimodal Accepts or produces more than text: images, audio, video The medium changes what a good prompt contains
The prompt is the model's only view of the task. Missing context is filled with the most typical answer.
Same prompt, different answer: output is probabilistic. Test a prompt more than once before judging it.
Generative AI Models that create new text, images, audio or code from patterns learned in training data.
Prompt The input given to a generative model: instructions, context, data and examples.
Token The unit a model reads and writes; a word or part of a word. A model gives a confident answer about an event after its training cutoff, with no web access. What is most likely?
A The answer is current B The model searched its memory index C The answer may be invented or outdated D The model refuses by design
Which setting makes output more varied between runs?
A Higher temperature B Lower temperature C Shorter context window D Earlier training cutoff
To pass Why prompt engineering
Scope, specificity, context
Text, image and sound prompts
Evaluating and improving prompts
Format Proctored exam + written task
Pass Competent on both Zero-shot prompts give no examples; few-shot prompts include worked examples so the model learns the pattern in context. Count the examples to tell them apart.
Chain-of-thought prompting asks for intermediate reasoning steps before the answer; zero-shot chain-of-thought just adds "Let's think step by step."
Supervised learning trains on labeled examples, unsupervised learning on unlabeled data, and reinforcement learning through rewards and penalties for actions.
The course builds effective prompts on scope, specificity and context: set the task's boundaries, state exactly what you need, and supply the background.
A student who passed named the core skills as specificity, context-setting, role assignment and iteration, plus how different models respond to prompt structure.
Objective and performance assessment Start studying Tue, Dec 8 Pre-assessment Thu, Jan 7 Objective assessment Tue, Jan 12 Submit the task Tue, Jan 12 Read the PA task and rubric first; plan prompt tests around each rubric row. Keep a prompt log as you test: version, change, output summary, score against criteria. Show iteration: a weak first prompt, the changes, and the measured improvement. Explain each change with a named technique: scope, context, few-shot, chain-of-thought, a prompt pattern. Cover each medium the task names: text, image, sound. Verify any facts or sources in AI output before they appear in the submission. Disclose AI use as the task instructions require; submit only original writing. For the OA: know the technique names, the parts of a prompt, AI limits, and responsible use. Learners create effective prompts with the consideration of scope, specificity, and context to elicit targeted information. Learners evaluate the efficacy of writing different prompts on research outcomes and improve the depth and quality of their analytical investigations. Learners evaluate the images, texts, and sound of the prompt and adjust prompts to output relevant results. Learners explain why prompt engineering is necessary. Sources Updated Nov 4, 2026
