The concern of copyright and data ownership in regard to Generative AI
In my previous blog post, I went through an overview of the 4 major concerns regarding generative AI (GenAI) from a business context. Today we start the journey of doing a slightly deeper dive into each of these items and how Cellaware Technologies is approaching these subjects to ensure that our products are safe for warehouse operations. Specifically, today we are going to dive into copyright and data ownership.
But first, we need to make a clear distinction between Large Language Models developed by companies' vs the products that they make with them; products such as OpenAI's ChatGPT and Dall-e, Anthropic's Claude or Google's Gemini.
Let's take OpenAI for Instance. OpenAI has been at the forefront of what is coined the "AI Arms Race" by many pundits. OpenAI, founded by Sam Altman, released the groundbreaking application called ChatGPT in November of 2022. ChatGPT was the culmination of many years of hard work by very intelligent people to develop a Generative Pre-trained Transformer (or GPT for Short) that could generate text based on a given prompt and could respond to the user in a conversational way. When you think GenAI, most people think ChatGPT, however, the engine that powered the original ChatGPT was a model called GPT-3.5 that was trained on billions, if not trillions of datapoints (more on that in another blog). This Model was developed and released as a finished product only knowing information up to a specific point in time.
Well, it was relatively well known at the time, and even more known now, that OpenAI was using information that was being inputted into ChatGPT to continue to train and fine-tune their model to perform better. The use of ChatGPT ranged from individuals trying to make a social media post to businesses and employees inputting proprietary information into chatbot. Obviously, this is very scary stuff. But as I mentioned, we need to make a clear distinction between ChatGPT and the GPT-3.5 model. Shortly after releasing ChatGPT, OpenAI also released an API to be able to interact with the base model so that companies and businesses can use it to develop their own applications (Enter Cellaware Technologies). This API leverages the already developed 3.5 Model, and while they did prioritize API users that were willing to commit training data to the model during the beta phase, by default, information translated to the API was not, (and still isn't) used to train the GPT Model. In fact, shortly after announcing this change in their privacy policy, OpenAI released a ChatGPT for Business soon to be followed by Microsoft releasing its Business CoPilot that makes those same guarantees.
Why is this distinction important to make? Cellaware Technologies has made the decision to use the best GPT models in existence and currently relies on a combination of GPT Producer Models and self-hosted open-source models. Unfortunately, to date, many of the open-source models do not perform to a level necessary to ensure consistent and reliable performance of our application. To ensure the safety of our client's data, Cellaware Technologies has as part of our policy that we will not opt into data sharing of customer inputs or customer data to train GPT Models for any vendor that we choose. This is a major reason why we have chosen to stay away from models produced by Meta and Google as they have yet to make this as an option for their API users. At Cellaware, we take the protection of our client's data extremely seriously and as we were building our GenAI solution, these thoughts were at the forefront of our minds. We want our customers to know that their data is safe in our hands.
At Cellaware, we love Generative AI and Warehousing and have created ChatWMS, the first ever GenAI solution for the warehouse industry. ChatWMS allows your warehouse operations leaders to have a conversation with the warehouse data, putting the power of data directly into the hands of the people managing the business.
We would love to show you how ChatWMS can benefit your operation. Schedule your free demo today!
Commentaires