Public ChatGPT vs Atlas Prometheus

You may be one of the many pondering over this question: How do I provide access to ChatGPT (or other LLM type of AI) while protecting your IP (intellectual property) and governing the results.

And it is clear, you are not alone, as we are seeing an big increase in enquiries from Atlas customers -and prospective customers - looking to understand how to deliver safe and useful approaches for their users to engage with AI, and therefore would like to understand more about Atlas Prometheus.

We launched Prometheus in preview – as an Atlas AI tool to demonstrate the powerful combination of Atlas with private, contextualized and governed interaction with a private LLM AI model (such as ChatGPT).

There are a number of considerations organizations should take into account, for their own protection, and for their end-users to take efficient advantage of AI in the flow of their work. I have listed those that I regard as the most important here.

• Provide the organization with an alternative to public AI tools, allowing the organization to “lock down” use of AI through clear policies, while offering a safe and governed enterprise-wide approach to end-users.
• Ensure that any prompt – from an app or an end user – does not go to the “public models” (such as OpenAI’s ChatGPT service), but is instead sent to a private copy of the LLM model (i.e. a copy of the ChatGPT model, hosted in the organization's own Azure subscription).
• Provide the user interface in a familiar, easily accessible and intuitive format to end-users.
• Help users by standardizing and contextualizing their prompts, in the context of their work, and in line with organizational policies.
• Govern who can use the tools using standard Azure Active Directory (Microsoft Entry ID) groups.
• Govern what happens to the LLM outputs (e.g. content has at least to go to Draft then through an approval workflow, before it can be published).

Atlas Prometheus demonstrates (link below) how this can be done. What I love most about it, is how it leverages the contextual awareness of Atlas, to generate draft content that is relevant - even if the prompt is quite generic or if I am not an expert in putting prompts together. 

Watch the 2.5 minute demonstration video in our resources area here.