TAMUDatathon Lite Challenges

TAMU Datathon Lite - Google Gemini Challenge

Team Members: Angela Yue, Jason Xiong, Matthew Shi, Brady Ji

Submission Guidelines: Writeup for the Google Gemini Challenge To participate, submit a detailed writeup of your most creative use of Google Gemini. Follow the guidelines below.

Detail Your Most Creative Use of Google Gemini

We used Gemini to guide us through many of the challenges in TD Lite. From brainstorming ideas to writing starter code to troubleshooting, Google Gemini was an extremely valuable tool that helped to boost our productivity. The most useful aspect of Google Gemini was definitely its content generation, as it was able to create the frameworks upon which we built off on to solve the problems presented in the CTF. For example, in the web scraping challenge, we were able to modify Gemini’s Selenium example in Python to fit our own needs and we managed to beat all the challenges in that category. There are various examples of this throughout the entire competition, as Gemini continually enabled us to work faster and more productively.

One challenge we had was when Gemini would occasionally forget its previously generated content when creating additional content, leading to confusion and falling into previous mistakes. In order to combat this, we tried to incorporate as much context as we could into our prompts in order to guarantee that we could maximize results, which ended up being a pretty effective solution.

Share Your Thought Process

We chose to use Google Gemini because of its accessible, advanced, and free AI model which is available to all students. It was a useful tool for debugging code and creating new content. Google Gemini also aided thorough our 8-hour hackathon process by providing consistent and accessible AI computing power throughout the various data-science related challenges. As opposed to some competitor models such as ChatGPT, Gemini created a paywall-free environment that allowed for continuous prompting, helping to democratize AI resources for those that may be financially challenged. As college students who do not yet have active income streams, we highly appreciated the lack of monetization from Gemini.

Through our process of using Gemini, there were a few errors Gemini made such as spelling “inputting” incorrectly, and a few code logic struggles, however, due to prompt engineering, we were able to utilize Gemini well and reach a final product to solve our problem. Gemini responded to prompt engineering quickly and relatively accurately, providing solutions to our code. For example, we needed to figure out why a decryption algorithm wasn’t working, and Gemini was able to pinpoint the location of the error, and provide key changes in easy to read, separate lines.

To experiment, we also asked Gemini to provide several solutions for a single problem, therefore allowing us to test out different ways to see which solution would be the most suitable for our project. Using Gemini allowed us to be both creative in our approach to the problem, and efficient in solving it. A specific example of this is asking Gemini to provide ideas on how to bargain with another chatbot. Gemini provided several different bargaining solutions when we provided it with prompt engineering techniques. It provided different labels (such as “The Emotional Appeal” and “The Value Proposition") and three examples for each label. This was more concise than the GPT model which only provided one solution for each label.

One of the main limitations we saw with Gemini was the reduced file compatibility with the attachments. We found that the general chat-bot and prompting feature was useful, but found it difficult to work with the restricted file types. It seems that Gemini works best with PDFs or .docx types, but cannot take in CSV files. Working with many data parsing, data validation, and web scraping, we believe that adding CSV compatibility would strongly improve the provided functionality of Google Gemini. Certain competitor LLMs have CSV file compatibility, and we believe that by including more file input types, Gemini could diversify its use cases, especially for tasks related to data science and data manipulation.

Explain Why Your Solution is Creative

Our most creative solutions for the Web Scraping Challenge was made possible through Gemini. The intended solution involved analyzing the HTML of the webpage in order to determine which player to eliminate through an automated Python script. However, we got the idea to reverse engineer the process of which player the program was going to choose next through analyzing the javascript code that ran the logic for the game. The issue was, we lacked the technical know-how to completely bring this idea into reality, and weren’t even entirely sure it was possible. Thankfully, we were able to convey our idea to Google Gemini, confirmed it was technically possible, and it walked us through the step-by-step process to reverse engineer a solution.

We believe that this is an instance of maximizing Generative AI’s potential: we didn’t just blindly follow Gemini’s instructions to build a cookie cutter solution for a problem, but use it as an amplifier to our creativity and were able to develop a unique and novel approach that would have otherwise remained a passing thought in our minds. This ability to bring seemingly unrealistic ideas into reality is what we believe to be the best use for Gemini.