Good at
Brainstorming, explaining, restructuring, summarising supplied text, drafting, and generating starting-point code.
Practical AI literacy
Get useful results from an LLM without trusting it blindly.
Large language models can help you explore ideas, explain concepts, draft text, and generate code. They can also produce convincing errors. This short guide shows a reliable workflow for using them with TechTahi.
1. What an LLM does
An LLM generates a response one piece of text at a time using patterns learned during training and the context you provide. It does not automatically know whether each statement is true, current, appropriate, or supported by a real source.
Needs checking
Facts, quotations, references, calculations, recent events, legal or health guidance, and code behaviour.
Your role
Set the goal, provide relevant context, test the result, correct mistakes, and take responsibility for the final work.
2. A simple prompt recipe
Clear and specific instructions usually work better than long, vague requests. Include only context that helps with the task, define constraints, and say exactly what the output should look like.
Goal
State the outcome you want in one direct sentence.
Context
Give the audience, purpose, inputs, and any background the model needs.
Constraints
Specify requirements, exclusions, length, technologies, tone, or accessibility needs.
Output and checks
Define the format and ask the model to check the result against your requirements.
Create a single-page study timer for first-year students.
Requirements:
- Use one self-contained HTML file with embedded CSS and JavaScript.
- Include 25-minute focus and 5-minute break modes.
- Provide Start, Pause, and Reset controls.
- Make it keyboard accessible and mobile friendly.
- Do not use external libraries or network requests.
Output only the complete HTML document, starting with <!DOCTYPE html>.
Before responding, check that every requirement is implemented.
Tip: If the first answer is weak, identify the exact problem and request one focused change. Iteration is normal; rephrasing and adding missing context can improve results.
3. Use the result with TechTahi
Keep the cycle small so errors are easy to find.
Describe
Use TechTahi’s Prompt Builder to describe the problem, users, behaviour, and constraints.
Generate and paste
Send the prompt to your chosen LLM. Paste its complete HTML response into Generated Output.
Preview and test
Try every control, test narrow screens, and inspect the browser console if something fails.
Refine
Describe what happened, what you expected, and any error message. Ask for a targeted fix.
Edit
Use TechTahi’s visual editor for text, layout, and style adjustments, then preview again.
Save or share
Save only after the page works as expected and you have reviewed its content and data use.
4. Check before you use or submit
Fluent wording is not evidence. Verification should match the risk: a colour suggestion needs little checking; a medical claim, assessment answer, or security decision needs authoritative evidence and qualified review.
- Confirm important claims with current, authoritative sources—not links invented by the model.
- Open every cited source and check that it exists and supports the claim.
- Recalculate numbers and test generated code with normal, empty, and unexpected inputs.
- Check that the response followed every requirement and did not add unwanted behaviour.
- Review language for bias, unsupported assumptions, and accessibility problems.
- Keep a record of meaningful AI assistance when your course, workplace, or publication requires disclosure.
5. Use AI responsibly
Protect information
Do not paste passwords, API keys, personal records, confidential material, unpublished research data, or identifiable student/client information into an AI service unless its approved data controls make that use appropriate.
Follow assessment rules
AI permission can differ by course and assessment. Check the assessment instructions, be transparent about permitted use, and ask your lecturer when unsure.
Respect authorship
Do not present unreviewed AI output as your expertise. Check originality, attribution, copyright requirements, and whether AI use must be acknowledged.
Keep human control
Do not automate consequential decisions without appropriate oversight. People affected by a decision should have suitable safeguards and a way to challenge errors.
Official further reading
Product features and institutional rules change. Use current official documentation when details matter.