Apple just blew the lid off of reasoning models by demonstrating that they break down and become highly inaccurate for many classes of complex problems. A must listen if you are applying AI to your business in any way !

Dive into the controversial limitations of cutting-edge AI! This episode explores groundbreaking research, including contributions from Apple scientists, revealing that Large Reasoning Models (LRMs) like those from OpenAI, Claude, and DeepSeek face a fundamental challenge: their ability to solve problems and reason collapses at high levels of complexity.

Gary Ambrosino unpacks a paper that investigates how problem complexity breaks down AI thinking, showing that even models designed for sophisticated reasoning struggle beyond a certain threshold. Discover the surprising findings from controlled puzzle environments, where researchers found AI reasoning performance can drop to zero accuracy and models even reduce their reasoning effort when faced with truly difficult tasks.

This discussion challenges prevailing assumptions about frontier AI capabilities and raises crucial questions about their true reasoning and generalizable problem-solving abilities. If you're interested in the science of AI, the limits of machine learning logic, and the future of artificial intelligence, this episode is a must-listen. Learn where current AI hits a wall [conversation history] and what it means for building more robust and reliable systems.