I learn, therefore I am

Large Language Models can help you do complex tasks effortlessly, tasks you were previously unable to perform without significant time and effort put in. However this does not mean you're actually learning something new. In fact, it can actually harm novice programmers^[1] (and in extent probably anyone learning anything new).

While using LLMs may allow you to do things faster than the old-fashioned way, it also often takes away the satisfaction of solving a problem. When you yourself work on a problem you might have to look up why the problem occurs, what other people have done to tackle the problem, and often adapt their solution to your unique case of the problem. This is especially true in software development.

On the other hand, using an LLM often feels like a battle where you're arguin against your screen to do something you have already mapped out in your mind and repeatedly steering it so it follows your thoughts and not the thoughts of the average training-data-supplier. When the solution is finally there it more often feels like survival instead of victory--at least in my own personal experience.

For my master's thesis research I will be attempting to refrain from relying on LLMs. They can occasionally be used as tools, but it should be like that: occasionally, meaning not as a part of your routine; and as tools, meaning that they should not replace your work and thoughts but complement it.

Learning new things is difficult. But that difficulty is the reason you learn. Learning skills the old-fashioned^[2] way might not be the path of least resistance, but the popularity of puzzles proves that humans enjoy even "useless" intellectual challenges, so why avoid them in your work?