My main takeaway from this week’s classes, simply put, is that the user’s input dictates the majority of the Large Language Model’s (LLM) output. If LLM users want to produce intended results, they need to be specific in their prompts. This includes detailing the audience, message intention, and other contextual matters. An important practice to create an optimal output is to use follow-up prompts. LLM’s are tools, not sentient beings, and must be instructed.

I really appreciated the quick lesson on why transformers matter and how they sequence words. This video can be found here. I found the theoretical visualization to be fascinating. In reality, though, it’s disheartening to imagine the energy required as data collection continues. This week, Adam brought up some great points regarding AI and environmental concerns. I find that notion to be exceedingly important as scholars continue to test, explore, and/or influence AI’s strengths and weaknesses.

Another ethical concern from this week’s classes was about how LLMs can perpetuate human biases at an accelerated rate. As discussed above, LLMs are dictated by their input, and this includes the input of data that they are built trained and trained on (Cheung). Unfortunately, humans hold all types of ignorant biases toward each other. Consequently, as stochastic parrots, LLMs will simply repeat harmful messages unless specifically regulated. Even worse, when prompted to generate images of people, LLM’s tend to depict typical stereotypes. For instance, they will show doctors as white males or dark-skinned Barbies as violent. In today’s age, all consumers must be exceedingly mindful of the products they consume. Especially with regard to products akin to LLM’s that provide quick results.

Cheung, Jon. “How AI Image Generators Make Bias Worse.” LIS, 21 Aug. 2023, lis.ac.uk/stories/how-ai-image-generators-make-bias-worse.

3Blue1Brown. “Transformers, the Tech behind LLMs | Deep Learning Chapter 5.” YouTube, 10 Sept. 2026, www.youtube.com/watch?v=CJ-BY-rQrVM.

Grammarly was used for spelling and grammar purposes.