It Takes a Pastor to Keep the Peace

An early criticism of ChatGPT and other LLMs was that ‘all they do is predict the next word’. Setting aside how revolutionary that is, one of the latest advances in technology is the ability for new models to reason through problems, as opposed to just returning the next most likely word. Recently at the NVIDIA annual technology conference, CEO Jenson Huang, presented a video with a practical example of how these models work.

It takes a pastor, to keep the peace. The reasoning model uses something called ‘Chain of Thought’, which just means it kept asking itself more questions until it reached the correct answer. This approach transforms AI from a fancy autocomplete into a thoughtful assistant capable of deeper insight. For example, when given a conflict scenario involving a pastor mediating between disputing parties, the model doesn't just predict the most likely next word. Instead, it assesses the context, considers possible motivations, and anticipates reactions—essentially thinking through the problem like a seasoned mediator would.

But how does this work practically? First, the model breaks down the problem into smaller, manageable questions. It evaluates each step individually, carefully building upon previous conclusions. By continuously questioning and verifying its own reasoning, the model significantly improves accuracy and usefulness. This advancement marks a shift in how we should think about AI. It's no longer a passive tool; it's becoming an active participant in problem-solving. For professionals, this means AI can now assist in tasks requiring judgment, nuance, and thoughtful analysis. This change opens practical possibilities for collaboration between humans and AI, especially in tasks that demand nuanced reasoning.