Recently, I ran into my old high school teacher. We talked about random things—I told him about college life, and he asked me if I’d ever used ChatGPT, and of course I said yes, because which student hasn’t? At this, he made a comment that struck me and will continue to do so, every time I remember it: “Sayang naman. You’re one of the better writers I’ve handled. Never sumakit ulo ko sa papers mo.” (“What a pity. You’re one of the better writers I’ve handled. I never got headaches reading your papers.”)
Aside from the satisfaction of being validated, I was also left to ponder on how ChatGPT has helped (or hindered) me. Where is the balance between utilizing technology to work efficiently but still cultivating my own skills and talents?
Before getting into the pros and cons of using generative AI tools, it is vital to understand exactly what they are and how they work. “GPT” in the name stands for Generative Pre-trained Transformer, which is the language model architecture that it runs on. ChatGPT functions in two phases: the data-gathering phase and the user interaction phase.
In the first phase, data is gathered via pre-training, which has two types: supervised and non-supervised. Supervised pre-training—which most AI projects used prior to the onset of generative AI systems like ChatGPT—is a process where the model is trained after labeled datasets.
Essentially, inputs are already associated with specific outputs. If you store a photo of an apple and label it “apple”, the next time a photo of an apple is detected, the model will be able to recognise that it is an apple. The more data that the model can learn from, the better.
Feed it enough photos of different types of apples, and it should be able to tell an apple is an apple, even if it is rotting.
Of course, there are limits to this type of pre-training, the most glaring being the fact that not all inputs can be anticipated by the humans “training” the model. Who would ever think to train a computer to respond to things like, “Explain Newton’s first law of motion using Ateneo boy conyo,” something that we very well know ChatGPT can do? (Don’t believe me? Check out the screenshot below, or better yet, try it yourself!)
The revolutionary technology behind ChatGPT that allows it to do that and so much more is called non-supervised pre-training, which essentially means that no specific labels are assigned to each input. Instead, the model is trained to learn the patterns and structures of the data. In the case of language modeling, the model is trained to understand syntax and semantics for it to be able to generate meaningful, conversational responses to user inputs. This system of pre-training saves developers time, allowing them to scale the model’s knowledge simply by increasing its reference database.
We’ve gone over the “P”, so now it’s time to break down the “T”. The transformer architecture is a type of neural network that is used for processing natural language data. It uses
“self-attention” to determine the importance of each word in a sequence. What’s special about this technology is that the model is able to pay attention to different parts of the input at the same time, rather than processing it linearly, so the model is able to respond more accurately.
The use of ChatGPT has numerous advantages. For one, it can streamline the research and writing process by swiftly generating ideas, aiding in overcoming writer’s block. ChatGPT’s rapid response time also ensures quick access to relevant information, enhancing productivity.
Meanwhile, for those lacking proficiency in grammar or writing style, it can serve as a valuable tool, providing guidance and improving the overall quality of their writing.
However, using ChatGPT also has its downsides. Despite the mounds of data it has access to, there always remains a risk of encountering inaccuracies or biases in the generated content, as even the training data may have its own inaccuracies and contradictions. Additionally, ChatGPT’s understanding is limited, which may result in responses that lack context or relevance to the user’s query. ChatGPT also cannot provide opinions or perform tasks requiring human judgment, limiting its utility in certain situations. Finally, excessive reliance on it may impede creativity, as we may become too reliant on its suggestions rather than honing our own creative skills through practice.
As a burnt out college junior, I’ve developed a habit of using ChatGPT, especially for submissions that I don’t particularly care to do. I’ve even taught my mom how to use it in the office. However, there is a balance between using it to enhance productivity and being too reliant on it. I personally strike that balance by being more intentional with my time, allocating more for creative endeavors so I can be more hands on with them. It’s also important to do further research when using it for academic purposes. Finally, much like with everything else in our lives, proper discernment is crucial in determining when it is and isn’t appropriate to use ChatGPT, because at the end of the day, all machines are simply byproducts of our own intellects.
Blog by Jana
References
Gewirtz, D. (2023). How does ChatGPT actually work? ZDNET. https://www.zdnet.com/article/how-does-chatgpt-work/
Guinness, H. (2023). How does ChatGPT work? Zapier. https://zapier.com/blog/how-does-chatgpt-work/
Ortiz, S. (2024). What is ChatGPT and why does it matter? Here’s what you need to know. ZDNET.
Ramponi, M. (2022). How ChatGPT actually works. AssemblyAI. https://www.assemblyai.com/blog/how-chatgpt-actually-works/
Ambidextr Media 
Leave a comment