Demystifying LLMs and Threats My Journey

As sophisticated General AI and Machine Learning technologies become increasingly entwined into everyday business operations, understanding their fundamental principles, how they work, and the associated security implications has never been more crucial. In this engaging seminar, we'll present a business-friendly overview of GenAI and Large Language Models (LLMs), shedding light on their practical security implications and risks, rather than focusing on futuristic applications.

We'll pull back the complex veil on LLMs, simplifying its operations without oversimplifying the technology. Through this, attendees will understand the risk of uncontrolled disclosure of Personally Identifiable Information (PII) using an LLM is carefully managed. We'll then explore various LLM deployment scenarios, namely Public LLMs, Private LLMs, LLMs as a service, and potential LLM abuse by malicious agents.

We aim to provide participants with practical knowledge, discuss common fears with data ingestion and extraction, and recommend actionable strategies.

[JakeKaldenbaugh]

Caleb, this is a great resource for those of us focused on understanding how LLMs can be deployed into enterprise contexts. One of the most impactful insights of this presentation is idea that there is no distinction between the control plane and the data plane and therefore no way to protect against manipulations of the control plane through the data plane. Great stuff.

[sanhananda]

Great insights Caleb. Very keen for the part 2 of this. Any ETA on when this might be out?

[rhll]

42:23 I think I didn't quite understand the purpose of the privilege LLM in the solution, why do we need another LLM just to transfer the data to the less privileged/quarantined one?