"AI Exposed" Part IV: The Inconvenient Truth

Stuart McClure and Alen Capalik close out their "AI Exposed" webinar series with a year-end review of 2025 and predictions for 2026, arguing that the economics and technology underpinning the AI boom don't hold up. They grade their prior predictions (hype outpacing reality, "picks and shovels" chipmakers winning, AI failing at abductive reasoning, the 90%-of-code claim falling flat), then dissect warning signs of a bubble, including circular AI revenue deals, model collapse in GPT-5/5.2, data poisoning of sources like PubMed, and vibe-coding disasters that delete databases. McClure frames the path forward as post-transformer, experiential "world models" that are small, specialized, and personal, with a human always in the loop, situating WethosAI's personal AI as the answer via the EPOCH framework and system-3 thinking. Live demos show XO Voice tailoring a quantum-mechanics explanation to the user's cognitive style and a Minecraft test where WethosAI agents collaboratively build one house in under six minutes while general OpenAI and Gemini agents produce three disconnected houses in about 25.