Introduction

Welcome to the website of SafeHaven, a social media platform without AI generated content.

What We Stand For

We are a small team of 3 based in Vienna, Austria currently creating a social media app, which only allows certified human generated data to be posted. Our goal is to create a haven from AI generated content and also be a middle man between AI companies and artists.

The Problem

Back when Large language models were still in training, most of the data used by companies to train AI was unlawfully taken from creators. Most creators affected by this, did not recieve a single cent for their content, which caused lawsuits such as Visual Artists vs. Stability AI, or NewYork Times against Microsoft and OpenAI. However the damage has been done and the “well has run dry” so to speak. AI companies have scraped the majority of data off the internet. This however is not necessarily a bad thing.

Why AI Companies Need Human Generated Data

One would think that with the generative AI we have today, AI can train itself off its own content. This could not be further from the truth. When training AI on its own synthetic data, companies began to realize that the small imperfections of AI content began to compound on each other, until what was generated was barely recognizable. This phenomenon is called Model Collapse and is detailed in this Oxford paper: ModelCollapse. In order to train AI there must be a base of human generated data. This leads us to our solution.

How We Should Move Foward

Now that AI companies have scraped most of the human generated data off the internet they need to get more, and they cannot simply scrape even more data off the internet as it is now polluted with synthetic data. So here is our compromise. We are creating a social media platform where creators can choose to lease out their data to big data companies and get paid for it, on top of the revenue they earn from the platform. Our platform will act as the middle man and ensure that creators are treated fairly and rewarded for their efforts, and that any and all content posted on our platform is verified as human generated.

C2PA

In order to filter out AI generated content from human generated content, we will implement a C2PA metadata filter. C2PA is metadata attached to videos, photos, and other media that certifies content as human generated. More details can be found at C2PA. While this system is far from perfect, we believe it is a step in the right direction. We will also implement a peer review system where viewers can flag content as AI generated for our team to review.

Our Business Model

  • SafeHaven is free to sign up, post, and view for everyone. We do not charge creators or viewers any subscription fees.
  • Our revenue comes from data leasing and Monetization. Creators can choose to opt in to our data leasing program, which allows AI companies to license their content for training purposes. When a deal is made, creators receive a cut of the licensing fee directly — SafeHaven takes a commission from the transaction to keep the platform running.
  • This means our interests are aligned with our creators. The more value your content has, the more you earn, and the more SafeHaven earns. We will never lease your data without your explicit opt-in, and you can withdraw from the program at any time.
  • For further information please refer to our license

Human Generated Content

The following images are examples of the human generated content SafeHaven wishes to protect.

Mohnfeld bei Grumbach in Sachsen Photo by Kora27, CC BY-SA 4.0, from Wikimedia Commons

Gurudongmar Lake, Sikkim, India Photo by Yoghya, CC BY-SA 4.0, from Wikimedia Commons

Crescent Moon next to Venus Photo by Dennis W. Asfour, CC BY-SA 4.0, from Wikimedia Commons

Morteratsch Glacier Ice Cave Photo by Roy Egloff, CC BY-SA 4.0, from Wikimedia Commons