Arial Black

Inspiration

Community leaders often have a large amount of inbound or large audience they ideally engage with often. This can lead to heavy social media usage and require a lot of critical thought. To help address this concern and allow them to respond to their members at scale, we built Arial!

01. One-of-a-kind Voice

With Arial, you can input your own personal tweets, long-form content, and even your favorite role models to generate a voice that reflects your own personality and interests.

02. Tailored Responses

By providing Arial with the answers to common or frequently asked questions, it will be able to create smart, tailored responses with your personality.

03. Reply at Scale

With your uniquely personalized voice, Arial can quickly and easily respond to a large volume of messages on your behalf, saving you time while engaging the community at scale.

How we built it

Arial is built with GPT3 and the Twitter API. By interpreting the tweets of who signed in, we can make an accurate profile of the person and then generate relevant tweet ideas using text-davinci-003

Challenges we ran into

Fine-tuning even with a lot of data wasn't as effective as prompt engineering. The Twitter API is a little tedious to use since it deprecated 1.1.

Accomplishments that we're proud of

We got the functional prototype fully working and most of the frontend done!

What we learned

prompt engineering is king for the short term, fine-tuning and leveraging RLHF as a reward model will help up the fidelity and quality over time for this problem space. Embedding and semantic search might be a good avenue for enforcing opinions from the community.

What's next for Arial Black

We want to tidy up the frontend and then implement a reward model through RLHF, and also meet with more customers to figure out the flywheel and pricing.

Built With

• gpt3
• nextauth
• nextjs
• nextui
• promptable
• twitter