AI tools (e.g. ChatGPT) in learning

Can I use ChatGPT (or other AI assistants) when studying at Turing College?

Short answer is: use, but don’t abuse.

Generally it is even recommended to use ChatGPT to learn and do routine tasks faster. Here are some life hacks to enhance your learning:

You can use ChatGPT to explain concepts in simple words: ask for analogy or to “explain it to me like I am 5.”
You can also ask it to explain some piece of code that is not obvious to you (e.g. from books, guides or documentation).
If you want to study some additional topics, you can ask ChatGPT to make a study plan or roadmap.
You can use ChatGPT instead of Google. Just ask your question in the most concrete way possible and there you go. But be aware of possible misinformation!

In this guide we will dive into Dos and Don'ts with ChatGPT as your personal assistant at Turing College.

Understand the limitations

Firstly, when using ChatGPT, it’s important to consider important limitations of such large language models. They hallucinate: mess up the facts, come up with something that doesn't exist, but almost always sound very confident. Let’s clarify: ChatGPT is not an expert you can trust 100 percent.

So if you have a difficult and important conceptual question such as “‎Is it true, that…?” it is usually still a good idea to consult with other learners, JTLs or STLs. Especially when, depending on different formulations, ChatGPT gives different answers to essentially the same question. Here is the way to initially fact check something:

Ask ChatGPT to explain why something is true.
Ask it to explain why the same thing is NOT true (ideally in a separate chat).
If ChatGPT responds to both questions without any doubts… Well, it’s probably better to consult a human on this topic.
If ChatGPT responds to question 1 or 2 with something like “Hey, actually, this is not true!”,‎ then you can more or less trust it on this topic.
Ask it for references/links to articles or resources that support its assertions/statements. Check those references! References can be hallucinated as well – you may be given links that don’t work or names of books that never existed.

How not to harm your learning with ChatGPT?

It is cool to use ChatGPT, it really accelerates a lot of things, but there is a great temptation to stop writing code and solve problems on your own. This is a trap! Here are some things you should NOT do with ChatGPT:

Python and SQL exercises. It’s ok to write some boilerplate code to solve your problem when you are working with data, but the purpose of code exercises is to learn how to code. It's an illusion that with the emergence of ChatGPT, this skill has become unnecessary: in order to create good prompts and debug the code written by ChatGPT, you need to be a good coder yourself!
- If you didn’t write the code yourself, you will probably struggle to find flaws/bugs/gaps in it.
Ask it a question without thinking on your own, especially when learning new topics. To really understand the topic, you need to try to solve the problem yourself, this is how active learning works. Of course, it is often worth using ChatGPT, especially if you know what you're doing and don't want to waste time. But when you are just studying a topic, start solving problems yourself, and only then ask ChatGPT for guidance.
Solving tests at the end of the parts. The point of these tests is to consolidate your neural connections and test yourself. Of course, you can ask ChatGPT for explanations, but only after you have passed the test.
- Remember: These tests are for self-assessment, and their score is not used for anything other than to help you reflect on your learning. Using any tools for these tests only gives you a false sense of progress.

Important:

Turing College encourages the responsible use of AI tools, recognizing their potential to enhance learning when used appropriately. We advocate for a balanced approach where these tools serve as complements to active learning and critical thinking. The misuse of AI tools, such as substituting personal effort completely with AI-generated solutions in projects is considered a breach of academic integrity. Such actions not only hinder personal growth but also compromise the values of honesty and integrity.

Prompts engineering: tips and good practices

Generally, a good prompt is concrete, clear and supplemented with context and examples. You can check some guides on how to construct good prompts, it’s an essential skill nowadays. Here is the article from freeCodeCamp on this topic.

When using ChatGPT as a code generator:

Specify the versions of the libraries you use. Code examples can also help ChatGPT to use the proper version. It can be difficult to debug generated code without this step
Remember that ChatGPT doesn’t always provide the most elegant and optimal solutions. But you can iteratively improve it, asking ChatGPT to make the code more concise and optimize it!
Always provide the code you have already written as context, it really helps ChatGPT to ‎understand‎ you better and write the code you like

The other very cool use-case to mention is ChatGPT as a debugger. You can brainstorm ideas about where your mistake is. You can feed stack traces to ChatGPT so it can help you understand what happened or why something failed.

How can you super-power your data analysis with ChatGPT?

Here is a very good quide from DataCamp, where using ChatGPT to analyze data is explained with a real example.

Here are some funny things you can do:

Initial brainstorm of ideas. Let’s say you have a dataset and you just don’t know what to start with. Provide the dataset description to ChatGPT and ask what interesting things you can do!
Plan the analysis. To structure your work, it is always good to have a clear plan. Make it with ChatGPT from the project description!
Feature engineering!!! You can provide a list of features to ChatGPT and ask to come up with some useful ratio features and other ideas. This is especially useful, when you work with a huge dataset with tons of features.

Data visualizations. Sometimes, to save your time reading the documentation, you can just describe the plot you want to achieve and ChatGPT will generate the code snippet for you. But beware that it doesn’t always work and you need to be at least familiar with the visualization library you use to debug it

Learner's Guide 2024