Navigating the Risks of Using LLM-Generated Information in Business-Critical Applications
The potential for Large Language Models (LLMs) to enhance the efficiency of business operations is undeniable. However, the allure of readily available LLM-generated information masks significant risks, especially when used in high-stakes business contexts. This panel session will explore the inherent limitations of LLMs and how inaccurate, conflicting or biased outputs can compromise decision-making or disrupt operations and lead to financial losses, reputational damage or even legal liabilities.
Panelists will describe examples of how using LLMs without appropriate safeguards can result in costly negative outcomes. Not only can LLMs generate inaccurate or biased information, but they are susceptible to prompt injection attacks, training data poisoning and vulnerabilities that can expose sensitive information. Panelists will also address the challenges of validating and auditing LLM-generated information, the lack of transparency in LLM reasoning processes, and the potential for security vulnerabilities.
The session will conclude with actionable strategies for mitigating these risks, including responsible deployment strategies and robust governance frameworks, emphasizing the importance of human oversight and validation mechanisms for AI outputs.
Stephen Collins is Principal Consultant at 1024tm, a boutique consulting firm that serves leading-edge networking, telecom and cloud infrastructure companies across multiple segments of the enterprise IT and service provider markets. His clients range from startups to billion-dollar businesses, with projects typically involving a mix of business and product strategy, go-to-market planning, thought leadership, product marketing, business development and sales enablement.
Stephen has a proven track record of successfully introducing innovative products into highly competitive, rapidly growing markets, and he brings clients four decades of operational experience in a wide range of executive, product management, engineering, consulting, industry analyst and advisory roles, working primarily with communications equipment manufacturers and telecom software vendors.
In recent years, as a result of numerous client engagements, Stephen has developed a specialization in network visibility, Big Data analytics, full-stack observability and the application of AI and machine learning to drive network automation and service orchestration in complex, mission-critical service provider and enterprise IT infrastructure. He has applied this expertise to a diverse range of projects in 5G network analytics (NWDAF), Open RAN Intelligent Controller (RIC) xApps and rApps, SD-WAN infrastructure, Internet visibility and cloud-native infrastructure observability.
Stephen was hired out of college by AT&T Bell Labs, where he worked in software development on T-carrier data communications systems. Leaving in search of a startup opportunity, he joined multi-protocol router pioneer Wellfleet Communications as a founding engineer, where he spent a hyperactive decade in the rapid expansion of the global internetworking market. Stephen then went on to co-found Spring Tide Networks, which developed an innovative IP service switch for service provider networks, culminating in an acquisition by Lucent Technologies for $1.5 billion.
As VP of marketing at Sonus Networks and later at Acme Packet, Stephen gained experience in service provider VoIP. At Tatara Systems, he was active in the formative years of the small cells market. At Active Broadband Networks, he was involved in applying software-defined networking to the development of a Broadband Network Gateway (BNG) router based on a white box switching platform.
Stephen has been a frequent blog contributor and speaker at industry conferences, and he has authored numerous articles for industry trade publications. He holds an M.S. in Computer, Information and Control engineering from the University of Michigan and a B.S. in Computer Systems Engineering, Summa Cum Laude, from the University of Massachusetts, Amherst.
Chris has been in our industry since before its inception (the lack of hair helping to identify this). His most recent projects have been focused within the aerospace, deception, deepfake, identity, cryptography, AI/AdversarialAI, and services sectors. Over the years, he’s founded or worked with numerous organizations specializing in human research, data intelligence, transportation, cryptography, and deception technologies.
These days he’s working on spreading risk, maturity, collaboration, and communication messaging across the industry. (Likely while coding his augmented EEG driven digital clone that’s monitoring his Internet usage, and tea and biscuit consumption!)
When not working he can be found in Eureka, Missouri charging round the countryside on a mountain bike, or hunkered down with the kids experimenting on ways to take over the planet. From an observability perspective he’s large, hairy, often wears a kilt, and can be found on stage with a cuppa tea in hand trying to explain to audiences why they must ask more questions before clicking life’s big red button.
