By Rohan D. ‘25

In November, OpenAI released ChatGPT, an artificial intelligence (AI) chatbot which promised clearly-written responses to a variety of user-inputted prompts and became a viral sensation on social media and in everyday conversation. This technology has incensed the academic and professional communities and provoked questions regarding authorship and plagiarism. While ChatGPT responds impressively to prompts, provides information in numerous domains, and conducts conversations with almost human-like capabilities, ChatGPT is a machine with weaknesses, such as the inconsistent and incorrect provision of information.

ChatGPT is an AI program, which means that it is a set of technologies that enable computers to execute a multitude of advanced functions, including the ability to understand, synthesize, and communicate in multiple mediums of information. Because of this, nascent AI needs to be trained. ChatGPT was trained with supervised learning and reinforcement learning. With supervised learning, OpenAI programmers taught the program by simulating chatbot conversations from both ends, playing both the user and the AI engine. After the supervised learning and as ChatGPT improved its performance, the programmers shifted to reinforcement learning. In reinforcement learning, the AI program generated multiple responses to a given prompt. Subsequently, the OpenAI programmers ranked the responses from best to worst.

With this training, ChatGPT developed many features rendering it the most sophisticated and versatile AI chatbot to date, according to Leon Zucchini of Curiosity AI. Its promise to write essays and stories, compose music, create computer programs, perform well on most tests, and write songs and poetry with just a few clicks of a mouse significantly improves upon previous chatbots. In addition, Ms. Nicole Sorger, the Upper School History Department Head, stated that ChatGPT could prove useful as a brainstorming tool in the classroom: “We can… use it as a teaching tool and say, ‘Type this question into ChatGPT if you’re looking for ideas about how to possibly approach this issue.’ The way I am looking at it…[as ]a calculator in math class… for getting some initial ideas or backdrop of a foundation for what can help you move forward…in answering a complex prompt.” While ChatGPT might prove useful in some academic settings, Ms. Sorger goes on to say, “But at the end of the day, it’s not going to give you the answer.”

Some observers believe that the impact of ChatGPT is limited because of its inherent technical limitations, which manifest themselves with factual errors. For instance, Ms. Kate Tomatis, the Upper School English Department Head, told the Gryphon Gazette in an exclusive interview that the English Department experimented with ChatGPT by inputting a Crystal English writing prompt on the book Purple Hibiscus, and it returned an unsatisfactory essay with quotes that weren’t even in the novel. She concluded, “For a student that doesn’t know any better and turns something in like that, it is going to be pretty obvious… So I think right now if your teacher is remotely paying attention, it would be hard to pull anything over on that, but I think the fear is what happens when this software gets more sophisticated.” Additionally, Ms. Sorger said, “The types of responses that were given were a decent starting point for brainstorming ideas on a given topic, but [they] were still pretty boiler-plate and didn’t have the type of critical thinking that… we are looking for as Crystal teachers in our assignments.” Furthermore, Ms. Emily Fawcett, an Upper School English Teacher, identified that the sonnets that were composed by ChatGPT were “very bad… I said, ‘Write me a sonnet about rain,’ and it was able to make a rhyming sonnet, but when I asked it to do a sonnet without stanza breaks, it was basically a prose poem… And it kept recycling cliches…[It] kind of hurt my soul.”

Nevertheless, like all technology, ChatGPT will continue to improve as the AI powering the chatbot continues to be trained. Rather, it improves. Returning to Ms. Sorger’s calculator analogy, what started as a technology that allowed one to add 1+1 has blossomed into a very advanced handheld tool that can solve systems of linear equations and multi-order polynomial equations. We have seen the development of software, such as Mathematica, that can do just about anything one would see in a high school math curriculum. Similarly, computers could only execute a few functions when they were first invented. However, now Artificial Intelligence that communicates with humans as real people exists. And these technologies will simply get more complex over time.

The implications of ChatGPT extend beyond the Crystal campus. Many people in the writing and journalism industries will eventually lose their jobs, as ChatGPT can generate content much faster than a human can. According to an Oxford study, it is estimated that 47% of jobs in the United States are at risk, including copywriters, customer service representatives, coders, and marketers (The Atlantic). Furthermore, tech experts have warned that flexible AI would threaten white-collar jobs because robots and AI are bound to replace skilled office jobs that were once considered immune to automation in the near future.

Nevertheless, not all is lost. ChatGPT may force jobs in these industries to specialize even more. After all, the AI software can’t think critically, generate completely new ideas, interview experts, or examine historical artifacts. Moreover, ChatGPT can aid companies in making them more efficient, especially in statistical analysis.

This presents a serious question that English and History departments are asking across the country: How do we limit plagiarism among our students? Ms. Tomatis stated, “If [ChatGPT] does get better, we will really have to move to more in-class writing where we shut down… everyone’s computer… and that could be the future within five years, but I don’t think we are there yet.” She continued, “We would have to do more in-class writing if we get to a point where we cannot trust [the students]… I would imagine that as this software evolves, we may have to move into a model where we are stricter about running everything that is submitted through these plagiarism systems like TurnItIn or, if it gets really bad, all in-class writing, or handwritten writing. That would be a bummer.” Ms. Sorger added, “[We can] design our prompts and our assignments in such a way that makes it impossible for ChatGPT to come up with a response that is as detailed as what we are looking for.” Ms. Julia Rubin, the Upper School Director of Teaching and Learning, also agreed, “The advantage to it is seeing that ‘this isn’t going to help me write a Crystal-level essay; I am better off relying on the tools that my teachers have provided me with and my own critical thinking skills.’”

Another way to test writing samples is a new technology named GPTZero. GPTZero serves to detect ChatGPT in writing with two parameters: “perplexity” and “burstiness.” One of the first articles on GPTZero as described on NPR stated, “Perplexity measures the complexity of text; if GPTZero is perplexed by the text, then it has high complexity and it’s more likely to be human-written. However, if the text is more familiar to the bot — because it’s been trained on such data — then it will have low complexity and therefore is more likely to be AI-generated.” GPTZero was created by Edward Tian, a 22-year-old undergraduate computer science student at Princeton University. Tian tweeted that, “there’s so much chatgpt hype going around. is this and that written by AI? we as humans deserve to know!”

We should expect the unexpected as ChatGPT continues to evolve into a human-like entity, affecting everything from homework assignments to half of the jobs in the United States. As ChatGPT continues to improve, new security measures and new jobs will have to emerge to match its rise. The only thing that can be certain right now is that this is just the beginning of this boundless technology.