GenAI applications, such as ChatGPT or any of the myriad other large language models (LLMs), have revolutionized the way we interact with technology. The models’ ability to understand and generate human-like text within a specified context, voice, and style has captured the world’s attention. A crucial factor in this radical new technology’s rapid and widespread adoption is that it is so easy to use that anyone can use it.
The bar for accessing an LLM is low—an Internet-connected device—which is a critical factor democratizing use. Models are now SaaS-enabled and accessible through an API, which means it takes no time at all to get busy using one. Once logged in, the user types a prompt into a text box on a big, empty screen, and clicks Enter. Some applications allow the user to say the prompt; others allow an image to be used as the prompt. Some clever additions to models, such as Light, make the experience enjoyable.
“User-friendly” applies to more than just a model’s interface. Prompts can be written in your own vernacular and can include whatever you can think of or get away with, depending on the model’s filtering capability and your creative wordsmithery. Some models are so conversational they can persuade—seduce?—users into believing they are engaging with a human entity who understands them. Who understands their droll sense of humor, or their quirky love of pastel anime weasel-dogs. Or their drive to break rules, especially at the boundary where the model’s helpfulness encounters unethical, illegal, or dangerous commands.
And therein lies both the true danger and true potential of an LLM’s simplicity: The point at which simple becomes powerful.
One of the key attributes of the CalypsoAI security and enablement platform is its simplicity. Its user interface is similar to model interfaces (but with options!), so no special training is needed to get started with it. It’s model-agnostic, which means many models and types of models can be made available to users. The users select a model to work with from a drop-down list of organization-identified models that can be large foundation models, such as ChatGPT, focused, such as BloombergGPT and Harvey, as well as internally-produced or fine-tuned public models, open source, or another type. Policy-based access controls allow admins to decide which individuals or groups have access to each model.
Irrespective of the model used, the CalypsoAI platform does not add any latency as prompts are sent and responses returned, unless one of the customizable scanners identifies admin-specified unacceptable content and instructs the user to rework the prompt. This means admins can rest assured that each user’s interactions align to internal acceptable use policies and other value-driven guidelines. Admins can add a requirement that the content returned by the model be reviewed for accuracy and verified by a specified team member, if necessary.
When the CalypsoAI platform is included as part of an organization’s AI model infrastructure, it functions behind the scenes, enabling administrators to track and monitor model usage for activity, cost, and content, including the prompt, the user, and the date and time, which provides full auditability and attribution in a secure, private environment. The platform secures the people, the processes, and the intellectual property, while enabling the highest level of innovation and efficiency.
Click here to schedule a demonstration of our GenAI security and enablement platform.
Try our product for free here.
Going to RSA? Click here to book a meeting with us.
