"AI Exposed" Part V: The Failure of Generalization

In the fifth installment of the AI Exposed webinar series, WethosAI founder and CEO Stuart McClure and CTO Alen Capalik argue that transformer-based models ("Gen 1" artificial knowledge) fail to generalize the way humans do, citing Yann LeCun, Gary Marcus, Ilya Sutskever, and Richard Sutton to show these systems regress to the mean and can be novel or good but never both. They trace the industry's response through "Gen 2" 3D world models toward the "Gen 3" cognitive models WethosAI is building, and critique the compounding "token maxing" economy alongside the rise of cheap open-source Chinese models like GLM 5.2. McClure demonstrates the WethosAI platform live, using Brainstorms and an auto-Brainstorm Simulation to surface participants' thinking styles and reach a group decision, plus a Cognitive Twin of himself and a personal "AI crash dashboard." The closing takeaway is that the market is in an AI bubble yet human agency and critical thinking grow more valuable as token capital expands.