IM 889: We Are the Neanderthals - Why LLMs Fail at Complex Reasoning
Summary
Patrick Hillman of Logical Intelligence discusses Kona, an energy-based reasoning model that uses physics to provide deterministic outputs for complex tasks where LLMs fail. He argues that LLMs mimic intelligence but lack true understanding, making them unreliable for critical infrastructure without heavy agent layers. Hillman also critiques frontier labs for poor crisis communication and suggests a regulatory "brain trust" to manage AI risks. The episode concludes with hosts marveling at the rapid improvement in visual generation capabilities of new models like Opus 5.5.
Listen ad-free on Castria