How should students use ChatGPT?
A look at the strengths and limitations of tech's newest sensation
*taps mic* this thing still work?
Been a while! Let’s talk ChatGPT. For anyone who has somehow avoided the news lately, ChatGPT is the chat bot created by Sam Altman’s company OpenAI. It is a type of large language model that uses deep learning to generate text based on a given prompt. It was trained on a large corpus of text data and uses that knowledge to predict the next word or phrase given the context of the prompt. ChatGPT uses a neural network called a Transformer to generate text and can be fine-tuned for specific tasks, such as answering questions or generating creative writing.
The last three sentences were from ChatGPT in response to the prompt “explain yourself in simple terms.”
Like many folks, I’ve spent the last month or so experimenting with ChatGPT in a range of scenarios, from minor productivity hacks, to correcting my extremely elementary coding chops, to having it spit out a positive review of Fyre Festival along with an argument for why Billy McFarland is a visionary. Going one step further than Google, ChatGPT can engage in dialogue and follow commands. It’s fun and the possibilities are seemingly endless.
There is a lot to discuss here. But as I am finishing my time at Georgetown, the discussion around ChatGPT in school is front and center. Ban it or adapt? And what are the limitations?
This Feels Different
When you think about truly transformational technologies coming to the mainstream, what comes to mind? You can only pick two.
I pick the internet and the iPhone.
I was born in 1996 and am very much a member of the “digital native” generation. I grew up never knowing a world without Google search. But I have to imagine that when the internet came to life and search engines exploded throughout the 90s, it was a “wow” moment. The entire history of life, the universe and everything, at your fingertips, on a personal computer? Pretty crazy, and still is! It changed the way the world searches and consumes information forever. Similarly, the launch of the iPhone in 2007 gave a powerful platform for productivity and creativity to hundreds of millions of people around the world, kickstarting a cycle of innovation for mobile computing.
Are there other technologies that belong on this list? Absolutely. But these are the two most prominent examples, in my opinion. With such broad societal implications, ChatGPT, and generative AI more broadly, feels like another “wow” moment.
What are its strengths?
1. Creating content.
Marketing copy, a song, an entire academic paper, adding a particular style to a piece of writing? ChatGPT can do it very well. Some teachers have even reported that ChatGPT writes better than many of their students.
2. Writing and debugging code.
While it can’t build and launch an entire app for you, ChatGPT is a powerful productivity tool for coders, able to write code snippets in seconds and highlight elusive errors. It can also manipulate and organize unstructured data. That said, you still need to know coding fundamentals for this to be useful.
3. Explaining complex topics.
“Explain quantum entanglement.” You once had to search Google, then scour the results for the simple explanation you were looking for. ChatGPT excels at quickly breaking down dense topics, no matter the prompt.
Weaknesses of ChatGPT
1. Chat GPT is often wrong.
It struggles with logic problems and often generates factually inaccurate—though convincing—writing. Remember that when you give ChatGPT a prompt, the model isn’t searching the internet. It is making a series of guesses, based off its training data, as to the most likely next word until it satisfies the prompt. Sometimes, it guesses wrong. Stack Overflow, a knowledge-sharing website for programmers, banned answers derived from ChatGPT because of how often they are wrong.
2. Bias is present.
The training data is taken from all writing produced by humanity from the dawn of recorded history to roughly 2021. That’s a lot of data. Humans have biases. Doing the math, human biases must be present in the output of ChatGPT. Furthermore, the engineers at OpenAI inject their own unconscious bias when tweaking model parameters.
3. Harmful use cases.
ChatGPT can generate convincing text that sounds true but is outright false. Misinformation is already rampant, so this will supercharge the spread of false and misleading content online.
ChatGPT in School
Educators at all levels are figuring out how to adjust to the reality of ChatGPT. AI writing an assignment differs slightly from classic plagiarism, where the work already exists elsewhere (e.g. copying a Wikipedia article word for word). And as previously mentioned, ChatGPT responds to specific prompts with impressive accuracy. For example, if a professor asks a student to write a paper discussing the main elements of US First Amendment law, ChatGPT is more than capable writing it for you.
But I left out one limitation of ChatGPT: complex reasoning and analysis. Sure, I can have it summarize the key points and court decisions of First Amendment law, but if I then ask it to weigh the US regime of free speech against that of the European Union, and apply that analysis to the current issue of mitigating misinformation, the chatbot will fail miserably. Score one for humans!
ChatGPT writes at the level of a competent high school senior. It’s very matter-of-fact, complexity is low, little variation in sentence length, and it is completely lacking a writer’s voice. Educators need to proceed with the understanding that the genie is out of the bottle. Students will use ChatGPT for school work in a range of applications, and they should. Banning the chatbot isn’t feasible and makes no sense. Professors should instead exploit its limitations. You know it is great at spitting out facts (until it’s wrong, that is), so instead, ask students to do some of the trademark reasoning and analysis us humans are known for.
Make assignments “ChatGPT proof.” If you get an assignment that is fact heavy and surface level, you’ll know why.
On a related note, check out this article about a Princeton computer science undergraduate that built a bot for detecting AI writing. The variables used are interesting: perplexity, and burstiness. Basically, is his algorithm perplexed by the writing, and is there a variation in sentence length and structure? If yes to both, an AI likely did not write it. This tool is the start of an entire ecosystem for distinguishing between AI and human work. Will this detection technology show up on websites and social media? Will it always matter that we know whether what we’re seeing is from AI or humans? Someday, will there be no difference? So many questions.
Final Thoughts
If you write tweets for a living or are a top executive at Google, you’re on notice. This is the first mainstream, broadly applicable AI tool to hit the market. ChatGPT reached 10 million daily users in just 40 days, outpacing even Instagram’s initial growth. This figure implies at least 20 million monthly users. Insane numbers by any standard, considering Instagram took nearly a year to cross 10 million users.
This initial growth will moderate at some point, but anyone comparing it to other platforms like Instagram are forgetting that the applications of ChatGPT are virtually limitless. Uses go beyond novelty into real productivity. Time will tell. What does ChatGPT have to say about its own future?
PROMPT: Write one paragraph on the future of ChatGPT.
The future of ChatGPT and other language models looks bright as advancements in AI technology continue to shape our world. As these models become more advanced, they will likely be integrated into a wider range of applications, from customer service to virtual personal assistants. Companies and organizations will likely turn to language models to automate repetitive tasks and improve overall efficiency, further driving the development of the technology. Furthermore, with increased investment in AI research and development, the accuracy and capabilities of language models like ChatGPT are likely to continue to improve, making them an increasingly valuable tool for businesses and individuals alike.
Well, there you have it.
Cheers,
Ryan (& ChatGPT)