a16z Podcast: EliseAI's Mario Martone on AI Reliability
The a16z MTS podcast released a new episode featuring Mario Martone, Head of Applied Research at EliseAI. The program focuses on how reliability standards must differ from simple benchmark scores when AI errors impact essential industries like housing or healthcare. Martone explains how EliseAI handles roughly 5 million phone calls a month and why LLMs aren't the right solution for every problem.
ImportanceMaterialEvidenceE2 unreplicatedWrite-upQuick
The MTS podcast under a16z published a new episode on October 18, 2026, featuring Mario Martone, Head of Applied Research at EliseAI. The core topic of this episode explores what reliability standards should look like when AI system failures no longer just mean a drop in benchmark scores, but directly affect users' housing or medical services.
According to the publisher's show notes, Martone discusses the challenges EliseAI faces in actual operations, including handling a scale of approximately 5 million phone calls per month. He further explains why Large Language Models (LLMs) are not the universal key to solving all technical problems, and details how the team combines language models, traditional machine learning, neural networks, and internally built evaluations to determine which technical solutions actually work in production environments.
Additionally, Martone shares how his background in theoretical physics over the past 15 years has shaped EliseAI's research culture. This scientific methodology is reflected in everything from testing new models against their own benchmarks to fine-tuning open-source models and automating increasingly complex workflows across housing and healthcare. The episode aims to convey the case that applying AI to the real world produces intellectual difficulties comparable to those encountered in string theory research.