Unicorns to the sun

For those who are committed to understanding more than superficial name-dropping on AI, I highly recommend Stephen Wolfram’s What is ChatGPT Doing and Why Does It Work? It is a fascinating excursion into the basics of how we think and express ourselves. While data-trained neural networks must crawl 100 billion words from the web to learn and estimate the probabilities with which sequences should occur, the human mind intrinsically “knows” the meaning of words and sentences.

"Unicorns fly to the sun" is grammatically correct, but we are supposed to know that unicorns can’t fly and, since Icarus, that excessively high-spirited people have their wings melted when getting too close to the sun. When I asked ChatGPT if unicorns can fly, I got an answer that some VC investors should bear in mind:"Though unicorns are mythical creatures, they are often depicted as being able to fly. In legends and stories, unicorns are portrayed with magical abilities, including the ability to soar through the skies. However, since they are not real, their ability to fly is entirely up to the imagination!"

This response highlights a fundamental aspect of how AI models operate—they do not possess intrinsic understanding but instead generate statistically probable sequences of words. Unlike human cognition, which builds meaning through abstraction, intuition, and lived experience, AI derives its “knowledge” from an extensive pattern-matching process. This explains why models like ChatGPT can construct coherent sentences while still lacking the deeper contextual awareness that humans apply instinctively.

Moreover, this limitation has direct implications for fields where decision-making relies on more than just syntactic accuracy. In areas such as venture capital, economics, and strategic forecasting, an over-reliance on AI-generated analysis without human judgment can lead to misplaced confidence in probabilistically derived conclusions. Just as an LLM cannot independently infer the impossibility of unicorns flying without prior human-encoded context, investors and decision-makers must remain critical of AI-driven insights, ensuring they differentiate between statistical likelihood and real-world feasibility.