Traditional ML vs LLMs: who can classify better?

This presentation evaluates the effectiveness of traditional predictive machine learning models versus generative Large Language Models (LLMs) for critical cybersecurity classification tasks. While traditional ML models are specifically designed to learn decision boundaries for identifying threats like phishing and network anomalies, LLMs excel at learning input distributions and generating data. By comparing the performance metrics of both approaches, the talk investigates whether the latent discriminative properties of modern LLMs can outperform the specialized predictive capabilities of traditional algorithms in securing systems.

Mar 4, 2026 5:30 PM — Mar 5, 2026 7:30 PM

unprompted Conference, March 2026

San Francisco, CA