**p-Interact: Applying GPT-3 to Interactive Fiction** (Lead: Andy Liu HMC '23)

• Language models like GPT-3 have become very good at generating shorter sequences of text, but lose coherency and consistency over longer stretches, such as fiction.
• Goal: apply GPT-3 towards writing interactive fiction (stories where the reader can interact and influence the outcome).
• Work will include training GPT-3 to generate fiction, encoding information within GPT-3 prompts to maintain coherency, and evaluating model outputs.
• Ideal for those interested in computational creativity, working with GPT-3, and more applied ML projects.

**p-TCM: Bridging Traditional Chinese Medicine & Modern Nutritional Science** (Lead: Aaron Xie CMC '24)

• Use classification machine learning algorithms to associate modern nutritional science properties with the traditional Chinese medicine hot and cold classifications
• Gain familiarity with common elements of ML pipeline (eg. Data exploration, data cleaning, visualization)
• Train models using Logistic Regression, SVM, KNN, Decision Tree and more; tune and compare their accuracy
• See how ML is used as an analysis tool in science research
• Possible opportunity to collaborate with researchers in the nutritional science field to write it up and submit to journal (depending on results and quality of analysis)

**p-Climate: Reconstructing Climate Signals from Natural Records** (Lead: Hannah Mandell PO '23)

• Use the skeletal and physical composition of natural records to peer into Earth's climate past
• Explore the efficacy of Neural Networks in a field that has long relied on linear statistical models
• Intersection of environmental science and artificial intelligence -- a rapidly expanding field of interest!
• Recreate Earth's historical temperature, so we can begin to predict and understand the future of climate

**p-Deepfake: Feature Extraction of Deepfake Images** (Lead: Danica Du HMC '23)

• Deepfakes are manipulated digital media where the likeness of one person in an image or video is replaced with that of another person — a growing problem in modern society because of how difficult it has become to detect with the naked eye!
• We’ll research feature extraction methods useful for deepfake image classification,
• Create a deepfake image dataset from existing benchmark deepfake video datasets,