Conclusion Feedback

Semester Reflection

This semester marked a significant shift in my approach to coding and project development. Unlike my first semester, which focused on foundational concepts, this term was all about integrating new and complex technologies, particularly various forms of artificial intelligence. The projects—which included a generative novel, a fanfiction-style chatbot, and an AI-trained Teachable Machine application—were far more challenging, but the opportunity to work with these tools allowed me to learn and work at a much faster pace, tackling codebases of a complexity I would not have attempted before.

Prompting

A major part of this learning was mastering the art of prompting Large Language Models (LLMs). Initially, I was unsure how specific my instructions needed to be; often getting generic or unhelpful results. I quickly learned to formulate my prompts with extreme precision, providing the LLMs with a clear understanding of the desired outcome. Consequently, this skill proved to be invaluable for both creative and technical tasks and taught me a key lesson: an AI is only as effective as the instructions it receives as well as the data it was trained on.

Hybrid Workflow

This semester also taught me that AI is not a universal solution. I quickly realised that for certain tasks like basic layout and formatting, it was much faster and more accurate to implement the changes by hand rather than spending time trying to prompt an LLM to do it for me. This hybrid approach—using AI for complex logic while handling simpler tasks manually—became my most efficient workflow. I even found it was advantageous to use more than one LLM. For example, on the generative novel project, using both Gemini and ChatGPT together led to a better overall structure and more cohesive concepts, since different models have distinct training focuses and strengths.

Generative AI

My experience with AI image models was particularly eye-opening. While they could create some impressive visuals, it was an almost impossible task to achieve any kind of character consistency across different images. The models struggled to keep a consistent face, body, or expression, which was frustrating for storytelling projects. Furthermore, the operational costs of these models are a significant factor, making them not very practical for long, creative projects. Nevertheless, my theoretical knowledge from class about how LLMs work helped me understand why some prompts were successful and why some tasks were simply too complex for them to handle.

Conclusion

All in all, this semester's curriculum was much tougher, but it was also far more engaging. The projects were not just about creating a generative story or a unique app; they were a huge lesson in creative collaboration and adapting to real-world limitations. This experience showed me that a thoughtful, hybrid approach—using AI for the general creative work and manual coding for the specific details—leads to a better and more organised final product. Due to the fact that the projects given were more complex and challenging than in the first semester, I spend much more time working on the individual tasks. Furthermore, I, personally, found these projects far more appealing and intriguing, and I enjoyed working on them immensely.