Inspiration

Many of us are curious about scientific work published in journals like Nature and Science. However, these research articles are difficult to understand for a common man. Even though someone is an expert in a field ABC, he can't understand and appreciate a paper from another field XYZ. They have to either Google every difficult term or find someone from XYZ field to help. However, the recent progress in natural language processing, deep learning and artificial intelligence in general has made it possible for a computer to do that on demand.

OpenAI is an AI research and deployment company which has opened access to their state-of-the-art natural language models (GPT-3). These models can understand english language, grammar and even software programs in Python, JavaScript. All they need is a curated prompt to perform the desired task.

One of the capabilities of GPT-3 is summarization. It can summarize a long paragraph in a few short sentences if instructed appropriately. Once we find the optimal instruction, we can use it develop online tools to help the masses.

AI generated summaries of scientific research can now happen at scale. Any researcher can now spread the word of their hard-work to the masses. Any common man can understand and appreciate the scientific progress.

What it does

It summarizes a long paragraph in a few short sentences.

How we built it

I built it using JavaScript. I used ReactJS to develop the frontend and NodeJS for the backend. The code is hosted on Vercel platform for free using their Hobby plan.

The backend uses OpenAI API to summarize a paragraph from the user. This API isn't free but new users get $18 worth of credit.

Challenges we ran into

Deciding the prompt for generating summaries is crucial. It can range from a simple "Summarize this" to "Summarize this for second-grade student".

To keep things simple, my app uses a fixed prompt.

Accomplishments that we're proud of

The first version was published in a few hours. Thanks to ease of deploying on the Vercel platform and OpenAI API, I was able to focus on the user experience and core business logic.

What we learned

  • Vercel and OpenAI API are quite easy to use
  • GPT-3 is quite good at summarizing popular scientific topics
  • GPT-3 can improve at interpreting cutting-edge research (eg. Quantum computing) which can have newly coined terms

What's next for Gistify

  • Make the web-app self-sustainable by monetizing (fees or donations)
  • Gather feedback on generated summaries for fine-tuning models for scientific domain
  • Develop a pipeline to summarize the entire paper
  • Publish summaries of highly cited papers on a blog
  • Give more options to users

Built With

Share this project:

Updates