"AI Exposed" Part II: The Hard Truth

Stuart McClure, CEO of WethosAI, and CTO Alen Capalik host the second installment of the AI Exposed webinar series, arguing that current autoregressive LLMs are genuinely useful but far from replacing humans. Through live demos they expose recurring failures: models can't reliably count words in their own responses, disagree on how many moons Saturn has, mishandle real-world physical and spatial reasoning, and struggle with abductive reasoning and long-horizon planning. They also examine deeper structural limits including the memory and context-window problem, model collapse from AI-trained-on-AI data, training-data bias and censorship, and the gap revealed by benchmarks like Humanity's Last Exam, pushing back on claims such as Anthropic's prediction that AI will soon write most code. The takeaway is that AI demands humans in the loop and system-two critical thinking, previewing WethosAI's concept of artificial individual intelligence as the path forward.