Rowan Insights
March 2024

Rowan Patents and Large Language Models: The Next Evolution

For the last two months, our articles have focused on Generative AI. We shared an overview of the feedback and input our clients have shared with us. And, we outlined what we’ve developed to date. For the next couple of articles, we’ll be looking into the future — what’s next on our drawing board for development. In this article, we focus on our thoughts and plans for employing various Language Models, the brains behind the Generative AI capabilities.

LLM Agnostic
As of this article, our Rowan AI Chat functionality integrates with Open AI’s GPT 4.0 Turbo model (gpt-4-0125-preview) via the OpenAI API. However, we are making progress on our goal of being LLM-agnostic. We have developed the ability to select from multiple models for any given prompt, making it possible for users to employ different models for different needs.

Local Models
Many of our customers have shared concerns over confidentiality and disclosure when using cloud-based, 3rd party models. While Rowan integrates with GPT 4 via their API where inputs are not used to train the model, we are looking to further reduce concern over such issues.

One method is using local models — models that are downloaded and reside on a user’s system where the content of the prompts never leaves that system.

As a desktop application that also runs locally, Rowan Patents is uniquely suited to integrate these local models. In fact, we already have. We are currently beta-testing three local models that have been integrated into Rowan Patents. We are gathering feedback on how they perform, both from the perspective of speed and quality of response. (If you’re interested in being part of this test and helping to shape the direction of Rowan Patents in this area, contact us)

BYO LLMs
Some of our clients have asked us about integrating a model that they are using or are considering. One of our corporate clients has asked us about integrating multiple models they employ across the company for different purposes.

Our plan supports clients bringing their own models to Rowan for integration. As mentioned above, we’ve already done work to support and switch between multiple models. Of course, the devil is in the details, and we would need to know more about the specific models our clients are using or plan to use in order to fully integrate these models. If your organization uses or plans to use its own version of a model, we would be happy to discuss the possible integration with Rowan Patents.