Learning vs. Efficiency: The Work-Study Student’s Dilemma in the Age of AI

Recently, I received an email from OpenAI informing me of yet another evolution of their model: ChatGPT was now better and could be integrated into my browser. This kind of email would have made me happy six months earlier, but today, it fuels the many questions that computer science students like me are asking themselves. In a world where the progress of models keeps accelerating, what is our future? Will the computer engineering studies I was proud to begin still make sense in tomorrow’s world? Will I still have value when highly capable models cost far less than I do? These questions, still unanswered for now, are a real source of concern for many of us. However, letting myself be discouraged by an uncertain future is not my style. Whatever it has in store for us, I am convinced that there will always be different ways to find value. But how can we acquire that value? Well, as always: by learning new skills. Nevertheless, a new variable has been added: AI. This assistant, improving day after day, is probably the biggest trap for modern learning. Since I am confronted with this trap, I decided to make it the main theme of my article in order to find solutions.

Before tackling the issue presented above in more detail, I would like to define learning as a two-step process. The first is the discovery of new concepts through reading, videos, or conferences. Then comes the second, which consists of integrating them through a repeated sequence of trial and error leading to their full understanding. With this foundation established, I can say that artificial intelligence, even if I seem to be demonizing it, is a valuable aid for the first step. It has never been so easy to approach a new concept. Are you interested in a subject? Ask it to provide you with books to learn more. Did you not fully understand a concept? It will summarize it for you. Do you want to learn a new programming language? Ask it to explain the theoretical concepts to you. All of this can be done for free through an online chatbot; not taking advantage of it would be a mistake. Moreover, with any subscription, you will also have access to an agent mode which, if used properly, will allow you to improve your learning abilities. Let us take the example of a developer starting a new job. Documentation is rarely well made, and the code is not necessarily commented. Getting started with understanding the code can therefore become a difficult and tedious task. In this context, agent mode will allow your model to browse the many files in order to answer each of your questions.

In short, artificial intelligence is a blessing when it comes to discovery. However, it also hides traps that we would all prefer to avoid.

As presented earlier, the second step of learning consists of a sequence of trial and error, with each attempt allowing us to deepen our understanding of the concept. Try, fail, understand. Try again, succeed. This process allows our brain to anchor knowledge so that it becomes capable of reproducing the desired action. But what happens when, instead of doing the task, we entrust it to an artificial intelligence? Let us run the experiment.

For a given task, we send a prompt, the model tries, then fails. With a second prompt, we correct it so that it tries again and then succeeds. In fact, it is exactly like our own learning process. Except that instead of learning, we are training a model. The worst part of this story is that we feel as if we have succeeded. The code appears perfect, it works, and we have completed our task much faster than expected. After this first experience, we feel as though we know and are productive. As a result, we continue, task after task, prompt after prompt, project after project. Until the moment when the masquerade is revealed. Our manager asks us to carry out a task that we had already completed on another project. As usual, we send a prompt to our favorite model, which unfortunately informs us that we no longer have any available credits. At that moment, we remain confident: “I have already done this kind of task, I will handle it by myself.” Then comes the moment to code. And then… nothing. Nothing comes. Coding seems so difficult, and the task appears insurmountable. We are nothing but impostors.

In summary, artificial intelligence amputates our learning process while lulling us into illusions about our own abilities. It gives us the feeling of being good while a dependency forms within us, thereby limiting our intellectual abilities. Our brain, no longer confronted with difficulty, becomes unable to think when that difficulty returns. In fact, the abusive use of artificial intelligence is a new form of brainrot.

This conclusion reveals a dilemma that I am confronted with every day. Should I prioritize learning or efficiency? As a work-study student in a company, I am both a learner and a producer of value. Except that unlike before, I am also my own referee. Moreover, depending on the context, formative tasks are now carried out by a model rather than by a work-study student. It therefore becomes essential to question ourselves before starting a new task. But how should we choose between efficiency and learning? Which elements are important or not? In other words, how can we acquire value while remaining efficient?

To begin with, we must recognize that the ability to produce lines of code is no longer essential for a developer. To gain value, we must therefore focus our learning on other areas of competence, especially those where models are still behind. That is, those where it is necessary to think in order to take a step back. Among these areas, I would first mention modeling. This includes the ability to separate a business problem into different entities that communicate with one another. In addition to making the code easier if it is done well, understanding it allows us to communicate with all professions, whether they are related to computer science or not. Since it also requires an ability to understand business problems, it is therefore an essential skill in today’s world or tomorrow’s. Other aspects such as architecture through patterns and code organization are also key skills, allowing projects to evolve properly over time, which is something our famous intelligence currently lacks. What must be remembered is that these theoretical aspects are the skills that will create our value. In addition to being poorly handled by the different artificial intelligences, they are also factors of efficiency. Indeed, an average developer will see their efficiency improved thanks to AI, while an excellent developer will see it multiplied by 10 or even 100.

To conclude, I would say that AI and the different models are a double-edged sword. Without any questioning of their use, we will fall into one of their many traps. However, by consolidating our theoretical foundations and learning to use these tools correctly, they can become a factor of improvement that will make all the difference compared with others. So rather than feeling sorry for ourselves, let us use this revolution to think bigger and propose projects of a new scale.