Part 9/12:
One of the persistent issues is AI confabulation—making up facts confidently. Shapiro points out ongoing efforts by organizations like Facebook AI with projects like Sphere, which aim to create knowledge graphs and databases that ground AI responses in verified truth. Yet, even with these systems, hallucinations and inaccuracies remain problematic, especially given that current models like ChatGPT lack real-time connectivity to external databases and operate solely based on training data within a rolling context window.