What stays the same?

When I woke up suddenly (perhaps because I had slept early last night), I began to think of creating a software product. It has been 4 months since I left a software company. Software is the automation of our daily tasks. What if I automate something I do daily, such as learning to write in English? As a non-native speaker, I have been struggling to articulate my thoughts in English words. Consequently, I am constantly learning English day by day—improving grammar, word choices, the use of compound sentences, among other things.

In this age of AI and the increasing success of their methods in natural language processing, it is unwise not to use them to enhance our learning. Many people use these LLMs to offload tasks such as coding, end-to-end researching, or doing an entire assignment, and I believe there is nothing wrong with this unless you want to excel in those fields. You can’t become an expert at programming by letting an LLM do all the coding for your project, nor will you be a good researcher by letting an LLM do all the work. I tried using LLMs this way, particularly for researching, and I found they made many mistakes by citing fake papers. Subsequently, I had to spend considerable hours checking whether they had done proper referencing. If we use LLMs this way, we have to be good at evaluation, and to be good at spotting errors, we should do most of the work ourselves. Of course, if the evaluation can be measured quantitatively and we don’t want to be experts in that task, we can offload it to an LLM, often reducing its exploration space, making it more likely to succeed.

When I watch YouTube and TikTok shorts, I am often reminded of the clip of Steve Jobs, where he said that when exploring what product to build, we can think in two ways: what is going to change, and what will stay the same. The latter is relatively easier compared to the former; it is very difficult to predict what will happen in five years compared to predicting what will remain the same over the next five or ten years.

Focusing on the latter perspective, what comes to mind today are our learning and teaching methods. I believe our methods will stay the same, and if we look closely, they have stayed the same for many decades. Technology hasn’t changed our capability for learning. We still have to go through this continuous process of getting feedback on our work and accordingly incorporating it into the “wires” of our brain. Even if we can generate 10 pages of paragraphs in a second using ChatGPT, we cannot increase our reading speed from one page per minute to ten pages per minute. Unless we evolve into different creatures with exceptional brains, it seems our learning methods will remain the same.

With these thoughts in mind and a constant urge to become a better writer (and communicator), I have been using LLMs (Gemini, ChatGPT, etc.) in the following way: Whenever I finish my writing, I prompt it to spot grammar issues and ask for suggestions for improvement. In the past, I used to prompt it to rephrase my writing, but that way it often produced the same kind of tone, losing integral aspects of my writing, including my personal style and tone. I believe each person has a unique style or tone of thought which is reflected in their writing.

When rephrasing, the problem is not only losing one’s tone but also missing important steps in learning to write. Every time I rephrase my words, I will never be able to fix my writing without my laptop or phone. Although I am online most of the time, I still want to enhance my writing skills and become proficient enough to effectively communicate my thoughts into words.

I believe the automation of writing-skill development is a valuable product due to its advantages in our schools. If LLMs can suggest words and fix grammar better than a primary English teacher, why should children wait an entire weekend for feedback rather than receiving immediate feedback? Why should I wait for my nerdy literary colleague to add "Contrastly" to the second sentence of the third paragraph of my research manuscript?

However, giving children (or learners) access to LLMs can lead to the temptation of skipping the learning process. After prompting "Give me suggestions for words in the following essay" and receiving hundreds of suggestions, making edits by reading them one by one is a lot of work. It becomes tempting to finish in a few minutes by prompting "Give me the whole essay with feedback mentioned above" and then watching one’s favorite show. We shouldn’t give this sledgehammer to learners and children because, in the end, they won’t achieve the learning outcomes they desire. This motivates me to build a different kind of chatbot which restricts learners from using this hammer, hopefully helping them become better writers.