THE WHY

For years important contributions to science, technology, and society in general have gone under-appreciated due to improper recognition, attribution, and record-keeping systems. With the growth of web3 and the decentralization of society, we need better systems to track and reward valuable contributions on all scopes, so we can incentivize continued work in these areas. To make tangible improvements has required:

Parsing massive amount of historical data and synthesize contributions
Storing a permanent, tamper-proof record of contributions

THE SOLUTION

Behold, ImpactGPT, a system that leverages GPT language models and the Hypercerts primitive (https://testnet.hypercerts.org/) to mint on-chain records of impactful contributions to science, technology, and society.

With this system we can scalably train AI models to identify important breakthroughs, so these models can make even the most brilliant & creative of humanity obsolete.

HOW IT WORKS

ImpactGPT takes prompts from users on a field of interest and uses a GPT3 model to generate a JSON object containing top contributors and relevant metadata (description of contribution, dates), as well as the DALL-E model to generate relevant imagery. This JSON / image data is then minted as Hypercerts on Optimism, creating a tamper-proof record of these contributors.

User flow:

Enter the platform at https://testnet.hypercerts.org/app/impactgpt
Enter prompt for an area of [science / technology / society] that you would like to recognize
Mint valuable generated hypercerts
[Future implementation] Evaluate the impact of generated hypercerts

WHAT WE HACKED TOGETHER

Before the hackathon, the Hypercerts Dapp already had the ability to mint hypercerts via manual form input (via /app/create).

During the hackathon, we created a new page (/app/impactgpt), where we added new OpenAI integrations. We crafted AI prompts to retrieve top contributions to a field and parse it into Hypercert metadata. We then prompted DALL-E to generate images for the NFT. To mint a hypercert, we just routed users to a pre-filled form on the pre-existing create page.

To make this project happen, we took advantage of the following services:

Plasmic no-code visual app builder
The Hypercerts protocol deployed on Goerli and Optimism
OpenAI GPT3 language model
OpenAI DALL-E image model

Note that this entire project was implemented using no-code tools, which is why we linked to the Plasmic Studio project, rather than a GitHub repository. Anything in the `/app/impactgpt` page was created during the hackathon.