the westminster news
Published by the students of Westminster School
By Wills Erda ’24
There is no denying that we live in a digital world. For many of us, the first and last thing we see every day is the glow of a phone screen. With all the wonders of technology that surround us, it can seem difficult to find something new to be surprising.
That is what we find in ChatGPT. For years now, we have been hearing about the rise of artificial intelligence (AI), and with this new chatbot from Open AI, we finally have a chance to play with it ourselves. ChatGPT (which stands for Generative Pre-trained Transformer) is a language model that uses billions of pages of input data to generate a response to a prompt using complex machine learning. These responses mimic the natural languages that we use every day.
Due to the enormous amount of information the language model has processed, it knows nearly everything and can answer almost any question. While the chatbot can write essays that could receive top marks in any AP English class here at Westminster, its real power lies in its concise explanation of complex topics. From history to math, biology, and the natural sciences, the bot can draw on its store of information and explain an array of topics, such as what the L'Hopital's rule is, how to draw a lewis structure, what are the causes of the Mexican American War, or what ATP is.
Users can also direct the chatbot's responses — from the sophisticated iambic pentameter to the vocabulary of a second grader. My favorite responses from the bot include “as a snobby NYT critic” or else “as an 80’s movie gangster”. Its language capabilities are also quite impressive being able to respond in a variety of languages such as Spanish, Chinese, and French. It can also translate better than popular translators such as google translate.
Perhaps even more powerful than its ability to work with human language is the chatbot’s ability to use computer code languages such as Python, Java, C++, C#, and more. While not great on large-scale coding projects, it can easily solve simple problems most efficiently. It has been a personal help in the development of my Backgammon AI, helping catch more basic errors in the code.
As I have hopefully been able to demonstrate, ChatGPT is a powerful and versatile tool. However, it is not without faults or limitations. First, because the model was only trained on data up to Q3 of 2021, it lacks knowledge of anything more recent. Furthermore, while the bot is often correct, it can produce misinformation with just as much confidence and persuasion as fact. The chatbot is a useful tool but is certainly yet to replace experts. Notably, one key issue that has plagued previous language models is absent. ChatGPT has also made large strides when it comes to social issues: it is neither racist nor supportive of Nazism, no matter how the users prompt it. ChatGPT is also trained not to give details on how to commit felonies, but crafty individuals can circumvent this by presenting a situation as a hypothetical or as part of a story. All this is to say that while Langue models and AI have come a very long way, there is still much to work on.
AI is fast becoming part everyone's lives, but it still presents many issues. One important thing to us here at Westminster is the possibility of cheating. Our tech office has already banned the site over the school internet over such concerns. With a bot that can write more compellingly than their students in seconds, teachers will need to adapt. Mr. Lawler for example in his AP English class had students write about “Their Eyes Were Watching God” as a comparison to “The Midcoast,” a recent book not in ChatGPT’s data set and therefore unusable.
While ChatGPT does present many cheating concerns, there are many ways for students to use it responsibly and ethically, such as using it for review, looking over essays, and receiving feedback. It is a tool that will only continue to grow in power and popularity and something we will all have to adapt to.