ISM Blogs

Over winter break, I continued to perfect my original work website by getting it to work on Render and improving the UI. Working on my original work has given me ideas on things I could possibly do for my final product, but I am still unsure of exactly what I want to make.

Over the past week in class, I have been planning my display board and working on it to make it eye-catching so that many people come over to my board. I am really excited to present it because I want to talk about how I trained my small AI model to reduce tokens of prompts for large AI models.

While working on my board, I have been thinking about how I need to get a mentor. I already have in mind my fourth interviewer, Ms. Adamaska, and I am waiting to contact her again. I emailed her last week, but I don't think university starts up until next week, so I will email her asking for a second interview. I want to ask some questions I had about my project and what I could improve, and also ask her if she would consider being my mentor.

Since she is a post-graduate researcher working on this topic, her insight could help me learn a lot about the field. I feel like green AI is a relatively new thing being researched right now, so there is a lot of room for me to learn and do stuff in it. In all, I am really excited for both presenting my board at the showcases and for learning more about my topic. I also really want to get a mentor and begin working on my final product because it is fast approaching.

Over Thanksgiving break and last week, I have been working on my original work and continuing to interview with professionals. Last week on Friday, I had an interview with Mr. Huang, a Data scientist for environmentally sustainable AI at Google. Through my interview, I learned a lot about how workers at Google are trying to lessen their energy consumption through various new techniques, which were mentioned in the article that I read.

I got a better understanding of how AI is growing to use more and more energy per day because of its unsustainable usage and creation. My idea of the token count in a prompt affecting energy consumption is very accurate, and the first layer of AI models. My interviewer told me my original work could definitely help people by reducing energy costs at the prompting stage, but also change their lifestyles. He brought up this point that really resonated with me: if we can change people's lifestyle and usage of AI, then consumption can inherently also be increased or reduced depending on the effects of the change.

Going deeper into my original work, I have coded and trained the AI models for my original work, and now I just face the task of creating the user interface as a platform for users to use my models. I plan to try and get a prototype of my app for the original work, but then develop it more during winter break and actually launch it later on. I faced a lot of bugs and mistakes while coding, but it also taught me a lot about AI, especially helped me understand at what level code really becomes AI, and how you can tell. That was one big question I didn't understand for a while, and I was really curious to find out, but I couldn't figure out how to differentiate code from AI. However, after training my models, I think I have better grasped what constitutes AI and what is just code in general.

The last 2 weeks of ISM, I have been continuing my research on creating an AI that can use reduced energy and be utilized to reduce the energy consumption of LLMs. Beyond my class research, we had a professional breakfast at Morning Bliss a couple of weeks ago, where I learned a lot about how to respectfully and politely eat. I also learned to be careful about what I order to make sure I enjoy the meal. I have learned a lot about professional etiquette and how important a skill that is in the professional realm.

I have continued searching for more people to interview with and plan on sending more emails soon. Now that I have learned more about my topic, I feel more prepared to speak with these professionals and really know what I am asking them and what I am trying to find out. I am fascinated by the topic and really want to make an impact. AI is using a lot of energy day by day, and I don't have the resources to create a completely different model with code that uses much less energy. As an alternative to this, I can build a much smaller model that requires much less energy and can be run on a personal laptop. I plan to start coding the base of the program this week and try to finish before the end of Thanksgiving break. Once I complete the code, I can work on building the user interface of the app and launch it on the Chrome Web Store for many people to use.

Last week in ISM, I had 2 new interviews, one with Mr. Goodfellow, the other with Mr. Vasireddy. In my interview with Mr. Goodfellow, I learned how he creates AI simulations of a fusion reactor for Google DeepMind. These simulations allow technical parts of the fusion reactor to be tested out, like the temperature and structure, so that clean, renewable energy can be easily created. He also believed my idea for my original work could definitely work and reduce energy, as the energy usage of prompts has been studied a lot recently. He proposed that one challenge I may need to overcome is actually measuring whether my AI model is reducing energy through its prompt reduction, as that is the main aspect of the whole project. I have been researching and working to better understand how I may go about solving that issue.

In my interview with Mr. Vasireddy, a PhD student at UTD, he proposed a different way to reduce the energy consumption of AI models. He explained to me that the amount of computations a model uses is the main energy consumer, so researchers have currently been trying to solve that issue through new coding architecture. I am going to take this new knowledge and implement it as well to reduce energy consumption even more. I now have a better understanding of how my prompt reducer can actually reduce energy consumption, which is by reducing the amount of computations the large model must undergo. Taking in all the information from both my interviews, I have figured out a way to grow my project to become even more impactful.

Both my interviewers provided me with multiple research articles and courses/videos, and I can review them to learn more about building my own AI model. Additionally, they have provided me with more people to contact. This week I will begin contacting more people for more interviews and hopefully find people with more experience in AI and energy consumption.

It's the start of a new quarter in ISM, and we have started to move away from just researching to contacting professionals in our field. I emailed and called 15 different people in AI, some more broadly, some more specifically related to deep learning in the environment. 2 people have already responded and said yes to an interview.

Mr. Cuman responded on October 17th, and I set up an interview with him on October 22nd at 4 pm. He has been working in the field for almost 40 years, according to his LinkedIn, and is now the president of Wentech. He seems to be interested in the business aspect of AI usage, so I am sure he can provide me insight into another aspect of AI's applications. I have 15 questions set to ask him, and hopefully they are not too basic. I really want to understand the more complex aspects of AI better to strengthen my foundation.

Mr. Goodfellow is the other professional I contacted who said yes to an interview. He is an AI research scientist and has worked at OpenAI, Apple, and Google, meaning he definitely has a lot of qualified experience. Since I want to become an AI researcher in the future, I think it will be very helpful for me to ask him questions and learn more about the job.

Last week on Wednesday I went to ISM symposium at Panther Creek High School, this was a completely new experience that I had never done before. At the symposium I met and talked to a lot of new people from ISM 1 and 2 students from other high schools to professionals in different fields. The first event at the symposium was the meet and greet with professionals, an informal way to just talk with people in different fields. I feel like I did well at getting myself out there and talked with about 3 people, but I could work on my ability to squeeze into other conversations because I was waiting too long for them to finish trying to be polite.

I learned surprising information from talking to these professionals both about my field and not, one explained to me how the field of education can start using more AI to help teachers enhance their teaching and have more time for it. From there at the ISM alumni breakout session I learned how ISM impacts people in the real world, and how it can help through college into a job. However, I faced a challenge in many of the breakout sessions during the Q & A portion. I wanted to ask something and facilitate the conversation but I couldn't come up with anything. I need to work on talking louder and being able to ask insightful questions so I can contribute to professional conversations and grow as a person.

Based on the great experience I had at the symposium I want to learn more about how AI can be used in different fields, but also how I can create a model that can help predict floods. My second interviewer told me how one nearby city faced a lot of damage from a flood because of how the city was built, so I want to learn how an AI could be designed to prevent a city design like that from happening, potentially reducing the damage that is caused.

In ISM, we have been preparing for the symposium at Panther Creek High School this Wednesday, where we are going to interview with and talk with professionals. This is a great opportunity to gain more experience in the professional realm and understand how to speak with them. I am excited to take part in the excellent opportunity where I can learn a lot about what different jobs look like in fields different from mine.

We have also completed a Research assessment where I read this article about using deep learning models to predict when a flood may occur and the damage that could result from it. The model they explained was called RAPFLO. It was really interesting to learn how it used various inputs and statistics to calculate the specific outputs of a flood. The article explained how a deep learning model would be able to calculate at a much lower computational energy and a lower cost than using a machine learning program because deep learning can use algorithms to predict complex data that is input. RAPFLO utilized hydrologic and hydraulic models to calculate the fluvial floods with extreme accuracy, but lower accuracy past 50 years into the future. All this information is very fascinating, and I may create something much simpler for my ISM project. It has given me new things to think about, as well as questions on how exactly the code would look and the algorithms that would be used for such a model.

Last week, we started and finished an annotated bibliography in ISM. Through my 6 sources, I gained a lot of insight into the basics of AI. I now understand that AI is all about the algorithms it is created to use, and based on that, it can find certain aspects of the data. However, I also learned issues about AI I never knew before: it has a massive carbon footprint, since to train the machine algorithms to accurately find certain information from the data, it must be fed a lot of data. I also learned that more powerful AI algorithms are made in such a way that most of the work to get from input to output is done completely invisible to the programmer, resulting in a need for a lot of trust. From my sources, I have been fascinated by deep learning the most, and I feel I could research more into that topic specifically. There are many ways deep learning can be used to help people. Some ideas I've found interesting so far are using it to combat climate change and reduce energy consumption.

Today, we are going to start preparing for our interviews at the ISM symposium next week. I am excited to meet professionals and gain more specific knowledge on the topic. Right now, I feel I have gotten a broad idea of AI, but I still need to learn more of the specifics, such as the programming side. I now know a lot more about how AI uses algorithms to form neural pathways similar to the human brain. I still want to learn more about how AI is programmed to do these incredible tasks.

I have learned so much about professional writing and researching in the past few weeks of ISM. I started this year knowing I wanted a career in AI, but I had no idea what that would look like until the career outlook assignment. I learned about one job in particular, an Artificial Intelligence research scientist, which sounds like an incredible job that seems perfect for me. Aside from that, I began researching for the Annotated Bibliography and gained much insight into the various types of AI. For example, today we have Special AI that can do certain tasks super quickly, but we do not have AI that is as smart or smarter than humans yet. I learned a lot about the basics of AI and different roles in its creation, but I want to go much deeper and really understand how it is created, such as the programming and mathematical side of the field.

In class, we also did professional resumes, where I learned you want to talk about everything relevant to your field that you have accomplished, because you never know what in your resume will catch someone's eye. Professionals look very quickly at resumes, so you have to work hard to make yours stand apart, and I learned that I need to gain more experience and leadership positions to really make my resume stand out. I want to continue adding experiences to my resume and learn even more about how algorithms are built and tested to make an AI function through ISM.