Fast, but at what cost?
I grew up in a time without large language models (LLMs) helping me learn. Because of that, I believe—or at least like to believe—that I learned how to learn. However, I do notice things are getting harder even with that background, because there is an expectation that we should be more productive with these LLMs. While I am able to do things faster with LLMs, it also costs me something: my learning. Take, for example, coding; nowadays, LLMs are far better than I am at it. To stay productive, the reasonable thing is to use LLMs. But by using them, I am also accumulating cognitive debt. I do not learn, and I become more detached from what is being made—my codebase. This is similar to technical debt, in that one common reason for it is that we want to be fast. Of course, wanting to be fast is not the only reason we accumulate these debts, but it is one of them. One of the things I value in life is learning. Without learning, what am I left with while being "productive" with LLMs? Thus, my plan now is to investigate how to learn while using LLMs as a support rather than moving fast without learning, because I do believe LLMs can be very helpful in helping me learn. In the old days, I always used to try to search for an alternative explanation when I did not understand the explanation given to me. Now, it is so easy; I can use an LLM to tailor the explanations to my level of understanding.
This post is written by a human and corrected by an LLM. I used the LLM like I used to use Grammarly to correct my text. So why write without an LLM? Well, because like the rest of the post, I want to get better at writing, reflect on what my beliefs are, and learn something from it.