Data will reach 40 Zetabytes of Data by 2020 – 44x in 11 Years

This means by now we generate a multitude of information every second – more than any human would be able to consume. In order to solve this problem we filter based on quality parameters. We grow brands like Media sites, like Frankfurter Allgemeine or Spiegel, that we trust to inform us with a good representation of the truth.

The point is, We believe the future can not be completely curated by humans. Humans are great at delivering high quality outputs and compressing information into communicable media.

However with exponential information growth it is quite challenging to expect humans with linear learning abilities to adapt.

Soo how could one potentially solve this problem? How can AI assist in media generation in the future?

Here is what we did:

The TL;DR is a prototype of a platform that will be partly-standard in the next 5 years

But it is a finished product with live data transformed to long-short-term-memory neural networks to create an interpolation from a broad selection of content sources to interpolate and create an artificial intelligence based news digest –

In short: We've created a news Site where every piece of content is AI produced.

What each and everyone of you can do now is access this platform and read articles the way AI currently perceives the world.

Potential

The potential is huge: For now the titles and articles are funny to read. It's an immersive experience that allows us to catch a little grasp on how the LSTM might work. We can imagine what the AI might have read in order to come up with these weird headlines and articles.

Tech Stack

Artificial Neural Networks for Article Generation: Char-RNN (LSTM Neural Networks) Training time ~8h GTX1080 CUDA ~1.250.000 Words

For headline generation Char-RNN (LSTM Neural Networks) Training time ~3h GTX1080 CUDA ~80.000 Words

Input Data for Article AI

Live Data from Content Pool ~1.250.000 Words

Input Data for Headline AI

Live Data from Google News Sites ~80.000 Words

Backend

PyTorch Python Flask API Locally hosted GPU Server

Frontend Flexbox based custom coded layout HTML5 CSS3+

The Team

Alexander Tonn

Deep Learning Expert

Maxim Lapis

Full Stack Developer

Kevin LaVerdiere

Product Owner

Daniel Seiler

UI/UX Designer

Tim Suchanek

GraphQL Engineer

Built With

Share this project:

Updates