PawnPwn

PawnPwn is an unconventional chess game where players describe moves for the AI to execute. Players attempt to creatively outsmart security checks, bypass traditional chess rules and gain an advantage

Comment

Inspiration

About our Game

PawnPwn is not just a normal chess game - it is a fusion of classic strategy and outsmarting artificial intelligence.

To make a move, you must describe it in words. The AI will interpret your commands and translate them into moves on the chessboard. The more creatively you articulate your strategy, the more intriguing the game becomes.

What sets PawnPwn apart? The possibility of "illegal" moves. If you're clever enough to outsmart our security checks, you can bend the traditional rules of chess, leading to a dynamic and constantly evolving game.

If you feel like your friends are too easy to beat by cheating, try your hand at beating Stockfish too!

How we built it

  • Google Gemini Pro LLM
  • React frontend + Socket.io backend

Challenges we ran into

Prompt Engineering

  • Engine being unable to fully comprehend the contextual knowledge given (ie chessboard state)
  • Output parsing
  • False negatives

Accomplishments that we're proud of

  • Exploring new LLMs outside of OpenAI's GPT
  • Completing a full application with only 2 team members present

What we learned

  • The non-deterministic nature of LLMs makes it difficult to handle edge cases and errors
  • Crafting prompts for an LLM to comprehend, process and respond according to a given structure requires a lot of finetuning

What's next for PawnPwn

  • Going multi-modal with speech to text movement
  • More stringent algorithms to prevent illegal moves
  • More variations in what moves can be parsed such as moving multiple pieces.
  • Having distinct and isolated models for each chess piece, enabling them to take on different characters/personalities and move independently
Share this project:

Updates