Jamf After Dark: Unpacking Jamf AI — The revenge of Spicy Clippy
Learn as co-hosts Kat Garbis and Sean Rabbitt, Jamf’s Matt Benyo, Manager of AI Initiatives and David Pryce-Compson, Director of Data Sciences talk about Jamf’s AI program and strategy.
Do you remember opening up Microsoft Word in the early 2000s, welcomed by the shapeshifting paperclip in the corner? This early digital assistant, Clippy, was a great resource for those who wanted to be bothered while working on a document. Or maybe you had some affection for its seeming desire to be helpful and entertaining.
With improvements to AI, these assistants have become much more powerful — even “spicy” as Jamf After Dark co-host Sean Rabbitt might say. At Jamf, we recently opened a public beta for our AI Assistant, ready to heat up your Jamf Account.
Jamf’s Matt Benyo, Manager of AI Initiatives, and David Pryce-Compson, Director of Data Sciences, both work with their teams to develop Jamf AI Assistant to make customer’s lives easier. To summarize, Benyo’s team is concerned with applications of generation AI, while Pryce-Compson’s team leverages their knowledge of language models and data to inform product development.
What is AI Assistant?
Benyo explains what AI Assistant is. Unveiled at JNUC 2024, AI Assistant is available in public beta.
The first iteration of AI Assistant will live in Jamf Account. After opting-in, users can access a chat feature where they can ask questions. The assistant will use knowledge from documentation, Jamf Nation posts and additional information from Jamf Support to answer questions. Customers will get more nuanced answers specific to their questions.
Eventually, AI Assistant will be in Jamf Pro, where customers can ask questions based on their specific environment, like “What version of macOS are devices on?” Customers will also be able to take actions, like creating smart groups and more.
Ensuring accuracy
Next, Pryce-Compson addresses how they make sure the assistant is delivering the correct information. Before Jamf AI Assistant was available in public beta, it was tested by the Jamf Support team. Their feedback proved invaluable, helping get data sources in the right shape to be queried and returned to the assistant. Behind the scenes, data is aggregated, ranked and scored, narrowed down, then checked for reasoning. The AI Assistant uses this information then summarizes this information for the user. This all happens quickly — on average 30 seconds between asking the questions and receiving highly relevant documents.
In the episode, Benyo breaks down how AI Assistant builds on Claude AI to feed trusted information to the user and prevent “hallucinations.” Listen for more details!
Balancing the power of AI with privacy
Privacy is a priority at Jamf; we are working with customers to understand what our boundaries should be. To summarize Benyo’s and Pryce-Compson’s points about this balance:
- The AI Assistant is opt-in, not enabled by default.
- The data is all in one place, and isn’t being sent off to anyone new.
- We’re investigating whether conversations should be preserved, and are leaning toward making them ephemeral to preserve privacy (pending resolution).
- Jamf has an internal council focused entirely on AI and its responsible use
- The data will not be used to train a language model
- Customer data will be treated as inference
Check out the episode for a deeper investigation about this balance, including:
- How Apple’s AI strategy informs Jamf’s
- Safeguards put in place to prevent vulnerabilities
- How the AI model deals with complex questions
Jamf’s long-term goals with AI
Benyo and Pryce-Compson discuss Jamf’s aspirations for the AI Assistant. In short:
- It will be available in all our different product portals, each with their own functionality.
- It will be a helpful tool with behavioral detections in Jamf Protect, helping admins know how to respond to the alert and providing rich information about the context.
- It will be used regularly by employees at Jamf.
Gain access to all the Jamf After Dark podcasts today!