Uptime used to mean reliability. But in the LLM era, five nines just means your liar is always available. Real reliability now includes correctness and that means probing models in real time with prompts that have known answers. When those slip, your delivery fabric has to reroute before customers find out. In this episode of Pop Goes the Stack, #F5's Lori MacVittie, Joel Moses, and returning guest Garland Moore dig into why availability isn’t enough anymore, and how research like “Get my drift? Catching LLM Task Drift with Activation Probes” shows where semantic health checks fit in the new definition of reliability. How do you keep #AI outputs accurate even when external data sources introduce bias, errors, or malicious prompts? Listen now to find out. Chapters: 00:00 Welcome to Pop Goes the Stack 00:37 Uptime in the LLM Era: Correctness and drift detection 03:09 Activation probing for semantic health 05:18 Real-world implications of task drift 06:44 Prompt injection through RAG data ingestion 08:38 Garbage in, garbage out: AI challenges 11:28 New standards for uptime 12:43 Hosted LLMs vs self-hosted LLMs 15:13 Passive vs. active drift detection 16:40 Activation probing: security impact 18:58 Key takeaways and learnings Read the paper, Get my drift? Catching LLM Task Drift with Activation Deltas: https://go.f5.net/vt7ad8rv Learn how you can stay ahead of the curve and keep your stack whole with additional insights on app security, multicloud, AI, and emerging tech: https://go.f5.net/av3lepur More about F5: https://go.f5.net/rw10snhw Read our blog: https://go.f5.net/suj4tix8 Follow us on LinkedIn: https://go.f5.net/juapq0r2

 91      1