The problem Gabby World solves

We aim to address a special intersection between two promising market opportunities: AI-native gaming and fully on-chain gaming.

1. AI-native gaming:

• Background: the rise of generative AI and experimental projects like SmallVille (the small town of 25 AI agents fully run on ChatGPT, created by Stanford University and Google) has unleashed endless possibility of building games completely generated and run by AI
• Opportunity: enabling a game to be built and run by generative AI massively lowers UGC barriers, allowing laymen to turn their imagination into game elements and potentially digital assets

2. Fully on-chain gaming:

• Background: pioneering projects like CryptoKitties, FOMO3D and Dark Forest have proven the possibility of creating blockchain-native games, which can provide composability and interoperability unmatchable by Web2 games and traditional game-fi’s
• Opportunity: similar to how de-fi provides a breeding ground for financial innovation that classical finance cannot offer, fully on-chain games maximise blockchain-native features and can create a new paradigm shift in future gaming experience

Challenges I ran into

Hurdle: off-chain LLM

• We rely on a large language model to run the game
• While all prompts are on-chain, the LLM itself is off-chain, so the prompts would have to go off-chain to be processed by the LLM, and then have the ouputs go on-chain again
• First issue it has is the high gas fee involved when outputs are going on-chain
• Second issue is that the outputs going on-chain cannot be fraud proven

Solution:

1. Using an oiracle to fraud-prove LLM outputs going on-chain: this is an ideal solution but mature solution isn't available in the market
2. Limiting LLM to 'storytelling' but not 'judging': as long as LLM outputs don't affect user win/loss, there's arguably low necessity for the fraud proof