The rise of large language models (LLMs) like ChatGPT has ignited a debate: how smart are these AI systems, really? While they can ace tests and even write research papers, they sometimes stumble on seemingly simple tasks. This raises a fundamental question: how do we measure the intelligence of an LLM? A new approach suggests separating AI intelligence into two categories: quantitative and qualitative. Quantitative intelligence measures an LLM's knowledge base—the vast information it can access and remix. Think of it as the breadth and depth of its learned facts. Current LLMs, trained on massive datasets, likely exceed any human in this area. But qualitative intelligence is different. It refers to reasoning, judgment, and drawing conclusions from new information. Can an LLM analyze a situation it’s never seen before and come up with a novel solution? This is where the real challenge lies. Researchers are exploring various techniques, like randomized controlled trials and crowd-sourced evaluations, to assess this more nuanced aspect of AI intelligence. However, there's still no standardized method for gauging qualitative capabilities, making it difficult to compare models and understand their true potential. While the size of LLMs and their training data continues to grow exponentially, their qualitative improvement seems more gradual. This begs the question: is simply rearranging massive amounts of data enough to replicate human-like reasoning? Even a hypothetical LLM trained on *all* human knowledge might still be limited by the very nature of its training data. If all it knows is human thought and language, can it ever truly transcend those boundaries? While LLMs may soon possess encyclopedic knowledge, the question of whether they can develop truly original insights remains. The search for accurate and comprehensive metrics for AI intelligence continues, and the answer could be key to unlocking the next stage of AI development.