You may have heard of, or used, generative AI tools (shortened to AI or AI tools from now on) like ChatGPT, Copilot or Gemini. These are powerful tools that we have seen both disrupt and assist the learning of trainees at CYF.

AI: A predictive word-generator

AI assistants, like those mentioned above, are often large language models (LLMs). LLMs are “trained” on huge amounts of data to predict what word should come next in a sentence. Because they can read and generate text data, it feels like you’re messaging a smart “person”, but it is just software that has an algorithm for guessing what words should follow your input ( prompt🧶🧶 promptThe input we give an AI assistant. ). Like a big “word generator”.

AI makes mistakes

Not only does AI tools predict the text to generate but it could be trained on out-of-date, limited or even wrong data. This means these tools can make mistakes or make up things that don’t exist ( hallucinations🧶🧶 hallucinationsWhen a large language model (LLM) perceives patterns or objects that are nonexistent, creating nonsensical or inaccurate outputs. ). For instance, a famous hallucination from 2024 was that many AI tools said there were two ‘r’s in the word strawberry.

AI hallucinations can be costly to a programmer.

It has been reported that AI can make up names of code packages which don’t exist, and direct a programmer to use the package in their code. This comes with a security risk: people can create malicious packages with names AI tends to generate often to exploit their “predictions” and get coders to import viruses into their code.

This means we should keep a critical eye on all output from AI tools.

Spotting issues with AI output is a key skill when using these tools. To thrive as a professional you need to have good judgement of what is a correct output. We can build this skill over time with feedback, discussion, insight from people with more experience than us, and constant questioning of “why” things are built the way they are.

Here are two ways to think critically about AI output:

1. Challenge AI output

We have observed that more experienced developers interact with AI very critically: they push back, ask for alternatives, get justification for the response from AI tooling.

Build this habit early. Don’t just use AI output without questioning it.

Ask AI for a different approach. Ask it to defend its answer. Ask it what could go wrong.

2. Always Verify

AI makes mistakes. It can produce code that looks correct but contains subtle errors, outdated practices, or security vulnerabilities. It can confidently agree with ideas that are wrong. Your job is always to check against documentation, against your own understanding, by running tests and more.

Look for verified sources:

• Ask AI to cite its sources and then check you trust where it found its information.
• Cross-check what you’re reading with real documentation straight from the creators of the library/language/software you’re using. A core coding skill.
• Use Slack to ask questions to your ITD peers and volunteers, what insights can you gain from the community.

Bad prompts can create bad AI output

AI can also be led to give incorrect information when a prompt isn’t accurate. Which means we need to ask good questions (or prompts).

As ITD learners, we are early in the CYF journey, we are likely still building our foundational software engineering knowledge. This means it is more likely we will ask AI inaccurate questions and AI will give misleading results. It’s something to be aware of and cautious about - the accuracy of the questions we ask is important.

For now, remember to challenge and verify what AI tools give us back. If we can’t explain it, or verify the information with our own fact-finding, we shouldn’t use it.

To read the full AI guidelines for education at CYF, please access the AI Use Guidelines for Trainees. This page will explain what acceptable AI use looks like for the ITD specifically.

Generally, AI might be a good tool for:

• Explaining tech concepts in plain language
• Answering “what does this mean?” questions
• Getting a starting point for research

You need to exercise CAUTION and avoid the following situations:

• Letting AI write something for you (e.g. your CV)
• Copying AI explanations without checking your understanding
• Just accepting the AI answer without exploring further

Our priority is building understanding

At CYF, we prioritise building understanding we can be employable in the tech industry. Employability in software engineering involves knowing how to debug broken code, find new information, and break down problems to find a solution. CYF mentors will try to help you without giving away the answers or writing the code for you so you learn how to overcome hurdles yourself and apply that skill in the workplace.

It can be easy for AI tools to do the hard work for us, e.g. solving the problem, give answers, fixing errors, or even writing emails for us. When AI does the work for us, we may think we’re making progress and learning when we’re not.

• We’re not understanding why we do certain things and what happens when we do them incorrectly.
• We’re not building skills that will make us employable.

We need to do that work ourselves, so that we build the deeper skills that will help us in our careers.

Why is it bad for AI to write for us?

Let’s take the example of a CV. We ask you to write your CV and keep it updated. This has the following knock on benefits to your skills:

• You are forced to think about your skills history and create your own narrative about your skills.
• By writing it yourself you will remember your CV content more, which helps with interview technique (recall of your skills).
• By writing it yourself you understand the structure and content of the CV, making it easier to make quick edits and updates.
• You improve your written English skills.

And what happens when AI writes a CV for us:

• You don’t build a mental map of your CV through structuring it yourself. This makes it harder to recall your own CV in interviews.
• Your CV lacks personalised tone of voice. CV comes out generic. Harder to catch the eye of recruiters, easier for recruiters to spot AI generation as it will sound like generated output, easier for the recruiter to discount your CV.
• Big risk of AI making things up about your background - remember AI hallucinates.
• You don’t build English language instincts.You don’t make your own mistakes so you don’t learn what is right or wrong. If you get a perfect instance of English generated in the first go, you are less likely to know what mistakes to correct in the future when you write your own words.

How do we use AI to build understanding?

1. Ask “why”, not “how”

Develop the habit of following up with questions that go deeper into understanding, e.g:

• “Why does this approach work?”
• “What are the alternatives?”
• “When would this not work?”

Avoid asking “how do I do X?” - this skips the understanding and goes straight to the answer.

2. Check if you’ve learnt something

There is a meaningful difference between “I used AI to help me understand this concept” and “I used AI to avoid engaging with this concept.”

The first makes you stronger. The second leaves a gap that will show itself later, often at the worst moment, like a job interview.

Ask yourself: Can you do this task next time without using AI?

If the answer is:

• Yes - you’ve learned the skill and can demonstrate it without AI.
• No - you didn’t understand and have to ask again, try to recap this question with less AI assistance.
• Depends - did you learn the research skill to help you find answers in the future without AI help?

CV writing: Human vs AI

Read three CVs

Let’s compare three CVs, one made by a human, and two made by ChatGPT using different kinds of prompts.

Review the CVs

We have a guide on how to write a great CV, use this to help assess what’s good or bad about the CVs.

We also have a automated CV feedback tool that can help give us feedback on the CV. We can even post a job description to the tool. All of the CVs above are for the job description as below:

Take a look at the CVs and answer the following questions:

• Which CV is more generic? What makes it generic?
• Which CV matches the job description the least and the most, why?
• If we take the human-written CV as the source of truth, did the AI generated CVs “hallucinate” any achievements?

We can see that the third CV over-exaggerated a lot of the experience and description of skills. It is full of “fluffy” language that aims to boost how skills sound on paper but comes across as generic and impersonal.

The second AI generated CV is even more generic, to the point of having almost no dates, no personal evidence of skills, no understanding of positive outcomes from past experience. It is as generic as can be.

The first human-written CV is shorter, but to the point. It backs up skills with evidence and doesn’t use fluffy language to look impressive. It only lacks a bit in experience which is more reflective of the experience of the person than the quality of writing. Because the applicant wrote this CV they could provide more accurate context and data to backup skills.

Questions

Question One

A CYF trainee is assigned coursework on a CYF course. The assignment is to build a homepage for a library that displays its twenty newest books. It requires that they use CSS Grid (a CSS feature used to display HTML elements in a grid format). They think that ChatGPT could help them complete this task.

Which of the prompts below is an appropriate use of AI by the trainee?

1. “Generate a webpage to display information about books in a grid format.”
2. “Use CSS Grid to build a webpage to display twenty digital books.”
3. “Build a homepage for a library website.”
4. “I want to learn more about CSS Grid. Please explain the core concepts simply with examples.”

Question Two

A CYF trainee prompts an AI tool to help them understand a complex CSS topic. The information the AI gives doesn’t sound right.

Which of the below might be appropriate reactions from the trainee?

1. Googling the topic and looking for reliable sources to clarify the point.
2. Trying to persuade the AI to check again.
3. Accepting this as the truth because the AI seemed to know what it was talking about.
4. Asking a volunteer whether the AI is right.
5. Asking another AI tool. 

Question Three

A CYF trainee balances full-time work and family responsibilities with studying at CYF. They are struggling to keep up with the course. In particular, they are struggling to understand some fundamental concepts.

Which of the below are appropriate actions that the trainee could take?

1. Asking volunteers for guidance at a session.
2. Giving an AI tool their assigned problems and asking it to solve them. They then copy and paste the output, making no edits.
3. Asking an AI tool to explain the concepts they find difficult to understand.
4. Attempting to solve the problems, then giving up and asking an AI tool to fix their code. Then copying and pasting the answer without understanding the changes it made.
5. Using Google to find articles on the topics they are struggling with. 
6. Copying another trainee’s answers.
7. Sending a direct message to an active volunteer.

What Should You Do?

1. Create a Google Doc

1. Create a Google Doc
2. Place your answers for each of the questions above into the doc
3. Add one to three sentences explaining why you chose the answers you did

2. Put your name in the document name

Include your given name or your family name in the title of the Google Doc.

3. Make your Google Doc public

Change the sharing settings of your Google Doc to “Anyone with the link can view”.

4. Submit the Google Doc link

Submit the link to the Google Doc in Step 3 on the CYF Course Platform.

A volunteer will review your submission and check that it meets the acceptance criteria. They will give you feedback on your submission and help you make any changes needed to meet the criteria.

Here are the criteria you need to meet:

• You have answered all three questions
• You have provided one to three sentences explaining each answer
• You have submitted the Google Doc link on the CYF Course Platform

Submission

You must do this on the CYF Course Platform.

1. Submit a link to your Google Doc

When you have a question, ask it.

When you have questions about coursework, ask in a workshop🧶🧶 workshopCode Your Future Workshops are meetings where you can get technical help. or in Slack🧶🧶 SlackSlack is messaging software - once you have downloaded Slack you can visit our Slack workspace to ask your questions.

You do not need to wait for a workshop to ask a question. You can ask questions in Slack at any time. We are here to help you understand and complete this course. Our goal is to help you learn.

Further Reading

You do not have to read these resources, but you might find them interesting or useful

• Hey programmers – is AI making us dumber?
• Why Prompting is the most underrated skill in the age of AI