Is there a (co-)pilot in the plane ?
Rédigé par Amélie Guillaumot & Romain Pialat
-
08 September 2025Can ChatGPT be used as an assistant in our daily work ? What are its limitations when it comes to more than just jokes ? What questions should we ask ourselves when using a tool such as the one offered by OpenAI ?
In this second part, we return mainly to the topic of IT development, which has caused quite a stir on the internet, detailing the possible uses of ChatGPT with examples.
So, is ChatGPT a co-pilot or a captain ?
Need a hand ?
Published on 04 April 2023
« I need a Python script that moves all files in the ‘data’ folder that contain the word ‘secret’ in their name to the ‘secret_data’ folder, which is at the same level as ‘data’, and encrypts them with a 16-byte key, which I will be prompted for. I need a confirmation mechanism for the key (for example, asking for it twice), and for the encryption algorithm I would like an AES-CBC-MAC ».
With that, your boss leaves your office and leaves you with your coffee, which you thought you would drink quietly while reading your emails. You did take a cryptography module during your studies, so you know what AES is, but what about CBC and MAC ? It was a long time ago. And anyway, you're not a developer either. You'll have to take a look at the Python documentation to find out which library to use for all this. Your coffee is cold, your emails are unopened, and this Monday morning is off to a great start.
Need a hand ?
Response given by ChatGPT when prompted with the text in quotation marks.
Both eagerly anticipated and feared by developers, ChatGPT is not limited to generating all kinds of text, but can also tackle scientific, mathematical, and especially computer science problems. By offering an interface similar to an Integrated Development Environment (IDE), the programs offered by AI use the codes of computer development, with its colors, indentations, and writing style.
We wonder to what extent this chatbot could become a developer's assistant, offering suggestions such as correcting careless mistakes or even writing entire sections of a program.
1. What role for ChatGPT ?
In general, ChatGPT is useful for quickly formulating coherent responses on a given topic. The same applies to the computer programs it suggests, provided you give sufficient detail in your request.
Don't feel like writing that script ? Or thinking about how to compare tables in a particular language ? No need to type keywords into a search engine and open five Stack Overflow tabs before finding the answer that comes closest to what you're looking for. ChatGPT will have already read everything for you and will offer you a solution tailored to your request.
In fact, some are wondering about the future of support forums, which will be irrevocably replaced by quick and easy contact with ChatGPT. For others, it is the future of AI responses that is uncertain : what will the content of future versions of ChatGPT look like, trained on an internet polluted with pre-chewed content produced by AI ?
ChatGPT has been conditioned here to translate into Spanish, as this was requested earlier in the discussion thread. It is therefore a joke to illustrate that it makes mistakes. However, it only makes mistakes if it is conditioned to do so. At least in this case. So it is a bad example, as well as being a bad joke.
Some tools, such as Copilot, already offer development assistance services, correcting errors, suggesting comments, etc. But when it comes to programming an entire function, the comparison between ChatGPT and the GitHub tool left no room for doubt, as Mathis Hammel showed in a thread where he proposed programming challenges to both tools.
Here we see that ChatGPT can seem very useful for occasional help, but it will not always be 100% accurate. In the case of an overly complex function or a lack of precision, it will probably end up suggesting obsolete methods or libraries that simply do not exist. Although it is regularly updated to correct these errors (particularly in mathematics), it is wise to remain cautious and accept that ChatGPT will not always be able to program everything for you.
So what are the limitations of this generative AI as a developer's assistant, and how can it best be used to improve your code and save time without taking risks ?
2. Possible uses
Apart from the very specific code seen above, ChatGPT can help you in many ways as a developer.
Several threads on Twitter have already summarized possible actions, often involving sending your code to the prompt, and therefore giving your potentially sensitive data to OpenAI. In this way, you can ask ChatGPT to clean up your code, add comments, give more explicit names to your variables and functions, translate from one language to another, etc., all of which are practical uses for quickly getting started on a new project.
However, such practices are not without risk. As discussed in article 2 of this series, the leakage of private data to OpenAI remains a major problem when you send your code to the ChatGPT prompt.
In response to a request for an explanation of the error « The truth value of an array with more than one element is ambiguous. Use a.any() or a.all() »
3. The joys of the chatbot ?
Thanks to its mastery of our language combined with that of programming languages, ChatGPT can simplify, explain, and even program. The methods it proposes are often accurate and work for specific programs or small functions. Just as it will struggle to write a long text without repeating itself, as evidenced by the many examples showing how to identify text generated by ChatGPT, generative AI will struggle to provide you with an entire program without making mistakes. An overly complex prompt will either return a mixture of different codes that appear to work from a distance but will actually require time to debug, or, more cautiously, a message stating that the request is too complicated, accompanied by a high-level version of the code to be written.
Apart from that, it is also important to know that ChatGPT does not incorporate all coding best practices into its programming. There may be a lack of optimization or misunderstandings of the request that can lead to data loss and, more worryingly, security breaches. It is necessary to check the code you are running and not to put ChatGPT's suggestions into “production” without validating them.
This generative AI is therefore a useful assistant in a creative and exploratory context, saving you time and helping you discover new things. It will not replace computer developers, but like other AI-based programming tools, it could profoundly change the way professionals work.
Summary
- Is there a (co-)pilot in the plane ?
- Lock, Stock and Two Smoking Bairrels
This article is part of a series on generative AI, which begins here !
Illustration - LINC avec StableDiffusion
