Speaker
Abstract
As large language models (LLM) become more prevalent in highly regulated industries, dealing with sensitive data and ensuring the security and ethical design of machine learning (ML) models is paramount. This session explores best practices for ensuring the security of LLM and ML models within these demanding environments. We will look at securing against common attacks such as prompt injection and sensitive information disclosure, what to do when your models hallucinate and the broader impact AI can bring to your organisation.
Attendees will gain valuable insights into common pitfalls inherent in the development of ML models and practical strategies to avoid them. By the end of the talk, participants will be equipped to navigate the nuanced landscape of LLM and Generative AI, armed with the knowledge to secure their models, design responsibly, and sidestep potential challenges.
Interview
Stefania Chaplin: The majority of my day is spent on working with enterprise organizations securing and optimizing their software delivery life cycles and improving technical productivity. A lot of my time is spent understanding the challenges associated with highly regulated industries, public and private sector. When not at a laptop, I enjoy speaking at conferences about Security and DevSecOps or running negotiation workshops for underrepresented groups.
Azhir Mahmood: I’m a Deep Learning Research Scientist at PhysicsX where everyday involves pushing the boundaries of what is possible with AI / Machine Learning. Being at the forefront of AI means I have to keep up to date with the latest research and then find innovative ways to solve problems for our clients.
At PhysicsX, I’m involved in every part of our product lifecycle from assessing our clients' problems, developing an innovative new solution and finally integrating and delivering the solution.
Prior to my current role, I spent a few years building a venture-backed AI start-up and then joined UCL as a Doctoral Candidate. During my time, I have been at the forefronts of both industry and research. Now I take all those learnings to help organizations scale and unlock new possibilities by integrating AI solutions.
Both Stefania and Azhir come from two distinct specialties: cyber security and AI. We discovered that Machine Learning Engineers / AI Scientists are not considering security vulnerabilities and those in security have not anticipated the sudden adoption of AI, security and compliance usually lags technological advancements. By bridging the gap between these two distinct yet complementary disciplines, we can ensure AI is adopted securely and responsibly within organizations.
Executives, Senior Managers and Technical Leaders are increasingly looking to integrate LLM and Generative AI into their organization. Those who haven't yet are increasingly being asked when and how their organization will adopt these technologies.
We provide a framework to let you confidently, securely, and responsibly adopt these technologies into your organization. Helping you to unlock new possibilities and accelerating your growth.
We hope that attendees will leave our session feeling inspired and confident in harnessing AI to discover fresh opportunities within their organization.
Here are three essential insights from this session:
1. Embrace best practices to ensure the security of your LLM / AI models.
2. Exercise responsible thinking in the design of your AI architecture.
3. Steer clear of prevalent pitfalls, including Model Bias and Data Leakage.
As a frequent attendee of QCon London, one of my favorite things is the conversations with attendees! QCon always attracts innovative leadership and learning how organizations intend to adopt and scale new practices is always an interesting conversation to have
Topics
QCon London 2024 is a three day conference for senior software engineers, architects and team leads. An international program committee of working engineers selects every session. Patterns and practices, not products and pitches.
From the same track
Monday 8 April
10:35 Windsor (5th Fl.) Session AI/ML Retrieval-Augmented Generation (RAG) Patterns and Best Practices Jay Alammar Director & Engineering Fellow @Cohere & Co-Author of "Hands-On Large Language Models" The rise of LLMs that coherently use language has led to an appetite to ground the generation of these models in facts and private collections of data. 11:45 Mountbatten (6th Fl.) Session AI/ML Navigating LLM Deployment: Tips, Tricks, and Techniques Meryem Arik Co-Founder and CEO @Doubleword (Previously TitanML), Recognized as a Technology Leader in Forbes 30 Under 30, Recovering Physicist Self-hosted Language Models are going to power the next generation of applications in critical industries like financial services, healthcare, and defence. 13:35 Mountbatten (6th Fl.) Session AI/ML Reach Next-Level Autonomy with LLM-Based AI Agents Tingyi Li Enterprise Solutions Architect @AWS Generative AI has emerged rapidly since the release of ChatGPT, yet the industry is still at its very early stage with unclear prospects and potential. 14:45 Mountbatten (6th Fl.) Session AI/ML LLM and Generative AI for Sensitive Data - Navigating Security, Responsibility, and Pitfalls in Highly Regulated Industries Stefania Chaplin, Azhir Mahmood As large language models (LLM) become more prevalent in highly regulated industries, dealing with sensitive data and ensuring the security and ethical design of machine learning (ML) models is paramount. 15:55 Whittle (3rd Fl.) Session AI/ML The AI Revolution Will Not Be Monopolized: How Open-Source Beats Economies of Scale, Even for LLMs Ines Montani Co-Founder & CEO @Explosion, Core Developer of spaCy With the latest advancements in Natural Language Processing and Large Language Models (LLMs), and big companies like OpenAI dominating the space, many people wonder: Are we heading further into a black box era with larger and larger models, obscured behind APIs controlled by big… 17:05 Fleming (3rd Fl.) Session AI/ML How Green is Green: LLMs to Understand Climate Disclosure at Scale Leo Browning First ML Engineer @ClimateAligned Assessment of the validity of climate finance claims requires a system that can handle significant variation in language, format, and structure present in climate and financial reporting documentation, and knowledge of the domain-specific language of climate science and finance.