Since ChatGPT was released, the possibility of using conversational AI has become main stream. ChatGPT was not the first chatbot, but it made the usage of chatbots simple and available to masses.

I have a concern: attribution. All these conversational bots use Language Learning Models (LLMs) internally to give you the answer. The effectiveness of the answer is based on the dataset that was used to train the model. The training data was created by Content Creators. But once the model is ready, no one reveals to you the input data that was used to train. The data need not be limited to text. It could be text, images, videos, etc.

This is a serious problem for content creators. They will not only lose recognition, they will also lose any opportunity to earn any revenue based on the content they created. For instance, if someone used posts from Substack to train their model, when their chatbot starts to answer questions, the authors in Substack will not receive any monitory compensation that they will otherwise receive through subscription.

I am not the first one to think about this problem. Stable Attribution is a project that provides attribution to the input images that were used to generate a particular image in Stable Diffusion. There also has been a backlash on why the method proposed by Stable Attribution project may fail.

There is a key difference between the Stable Attribution project and what I am concerned about. Stable Attribution is only attributing the input images that were used to generate the output image. I am concerned about all the images that were used to train the Stable Diffusion itself.