Yale experts explain ChatGPT
ChatGPT, the AI bot that drew over 1 million users soon after its launch, has taken the world of AI and education by storm.
Lizzie Conklin, Staff Illustrator
In November 2022, OpenAI, a San Francisco-based startup that works closely with Microsoft, launched their new AI chat bot, ChatGPT. As of today, ChatGPT has been used by over 500 million people worldwide.
GPT stands for “generative pre-training transformer” which is a coding architecture that is fine-tuned for specific tasks such as language translation, question-answering and text summarization. According to Tesca Fitzgerald, assistant professor of computer science and co-instructor of Computer Science 470: “Artificial Intelligence,” ChatGPT is the most advanced of a new generation of AI systems that can respond to users, generate coherent texts and even produce generated images based on a coded database of online content, digital books and various other media.
“Various beta versions of ChatGPT have existed before,” Fitzgerald said. “For the first time, anyone can access it.”
The previous versions of these “large-language models” were only available to researchers and used in academic settings. What makes ChatGPT different is its accessibility to anyone with internet access. To use ChatGPT, all users have to do is make an account that is tied to their email address. After joining, users can type various prompts and questions into the chat box, and the bot will respond with sentences that mirror human writing structure.
When the News asked ChatGPT to “explain ChatGPT to a five year old,” the AI responded with the following:
“ChatGPT is like a robot that can talk and understand what you say. It is very good at understanding words and sentences, and it can also make up new sentences that make sense. Just like how you can talk to your friends and tell them stories, ChatGPT can also have a conversation with you and tell you stories. It’s like a magic box that can understand and generate new sentences.”
In addition to asking requests, users can prompt ChatGPT to generate various texts, such as full-length essays and stories. Users could ask ChatGPT to write a 500-word essay about the American Revolution, and ChatGPT would be able to generate a unique response that mirrored human writing patterns.
However, the AI bot’s ability to generate these comprehensive and coherent texts has called into question the effects of the bot on education.
Professor of computer science Dragomir Radev noted that users must be “very careful when using ChatGPT” due to the “potential for false and misleading information.” According to Radev, although ChatGPT can generate comprehensive texts on various topics, its knowledge is only as strong as its dataset. If the dataset contains false or misleading information, ChatGPT will use that information to generate its response.
Radev also acknowledged the bot’s capabilities. He found the bot’s performance to be “amazing” noting that when he asked the bot to translate some English sentences into Bulgarian, it did so perfectly. Fitzgerald pictures ChatGPT’s role in education as a catalyst for change.
“This will force professors and teachers to re-evaluate how we assess our students’ knowledge of the things we teach,” Fitzgerald said. “I think ChatGPT will foster more critical thinking in the classroom.”
Fitzgerald recounted her own experience playing around with the bot. She asked the bot to generate an academic paper on human-robot interaction, a topic Fitzgerald is greatly familiar with.
ChatGPT generated a comprehensive academic paper, full of citations from authors that were also experts in the field. Fitzgerald even said that the citations were from authors that frequently worked together, and she thought she recognized some of the papers cited.
However, upon further investigation, she found that these citations were completely fake.
“The bot was smart enough to know that these professors had worked together in this area [and] smart enough to generate a title that made sense, but it wasn’t smart enough to actually pull information from existing papers,” Fitzgerald stated.
Fitzgerald’s experience demonstrates the limitations of ChatGPT and its dataset. However, professors have still acknowledged the bot in their classrooms.
Ashley Kwak ’24 told the news that her professor brought up ChatGPT on the first day of class, asking them not to use it.
“The first time a professor mentioned ChatGPT in class was on the first day of school this semester in relation to our writing assignments,” Kwak wrote to the News. “I was definitely surprised.”
At present, despite ChatGPT having a number of limitations, Radev anticipates that it will not be long before another model far surpasses ChatGPT.
“Other companies, such as IBM, Microsoft, Google, you.com, et cetera are all in the game, so [they] should expect major competition,” Radev said. “Government, legal and community regulations are likely to follow.”
ChatGPT was released on Nov. 30, 2022.