Inspiration

Inspired by the challenge of the Connections game from The New York Times, we wanted to explore how AI can replicate human-like categorization abilities and solve complex word association puzzles.

What it does

We built a Connections AI Solver which automates the gameplay of the Connections game, using natural language processing and LLM techniques to categorize 16 words into four distinct groups.

How we built it

We developed the solver using Python with natural language processing (NLP) methods, leveraging pre-trained models, and the OpenAI API to analyze word meanings and semantics. A Flask-based web app interfaces with the bot which allows us to engage in submissions. There were multiple methods that we tried throughout the project. For example, we began by taking a NLP approach by embedding words then using the K-Means algorithm supported by cosine similarity. On the other hand, we also attempted to characterize the words as embeddings, create a graph where each word is a node, and weigh the edges with their similarity. Lastly, we applied the Louvain algorithm as a means to cluster words with the limitation of four words per cluster.

Challenges we ran into

Ensuring accurate categorization with varying word associations was challenging. Balancing model performance and computation time to provide near-instant solutions for all game levels was also tough.

Accomplishments that we're proud of

We successfully created a program that can psuedo solve the Connections game, demonstrating the power of NLP and machine learning in real-time applications. The bot categorizes words with somewhat accuracy.

What we learned

We learned how to apply advanced machine learning models for semantic analysis, the importance of optimizing AI for user interaction, and how to improve the speed and efficiency of NLP systems.

What's next for Connections AI

We plan to refine the model's accuracy, expand its ability to handle more complex word groupings, and explore integrating other AI techniques to improve the game-solving experience.

Built With

Share this project:

Updates