dAlign.ai

Screenshots

Problem Statement

Problem:Aligning and steering Artificial Intelligence towards human valuesNeed to represent all of humanity, not just a select fewOnly 1 open-source alignment dataset of sufficient size and qualitySolution:Decentralized and permissionless dataset generationConsensus on controversial topics with no clear answerPitch presentation: https://docs.google.com/presentation/d/1lbBYT0qI8B3GCz5JhawMDibCXRfPZTRAyryzVGJl37M/edit?usp=sharingWebapp: Users are presented with simple tasks randomly. Such as creating a prompt, creating an answer to a prompt, rating the quality of a prompt, and selecting one of two answers to a prompt. Made for simplicity.Onboarding: A diverse population must go beyond the Web3 community, so the app is 100% free to use for users. No need any additional software (wallets). Enabled by the Account Abstraction stack from Safe Core (Auth, Protocol, and Relay kits) together with their partners Web3Auth and Gelato.Backend: A smart contract backend is deployed on the gnosis-chain so that you can generate datasets of sufficient size (100k promt-answer pairs) for pocket-change.Scaling: As the dataset grows in size, it is infeasble to process all of this data in frontend, subgraphs on The Graph enable the frontend to query smart contract data with GraphQL and work with large datasets without a centralized background.

Solution

Frontend: React, Safe Core Kits (Auth, Protocol, and Relayer), GraphQL queries to The Graph to simplify development and get ready for scale. On-boarding non Web3 users trough account abstraction and sponsoring all transactions.Smart Contract: Solidity, deployed on Gnosis (mainnet) to enable large datasets for next to nothing.