Musings

The current generation of AI (Large Language Models) are essentially very sophisticated pattern matchers trained on text. They excel at language tasks but struggle with precise mathematical reasoning. That’s why for forecasting, traditional time series models (ARIMA, Prophet, etc.) still outperform LLMs—they’re built specifically for that mathematical task. Similarly, for network optimization, algorithms like Mixed Integer Linear Programming will find mathematically optimal solutions that LLMs can only approximate. The magic happens when we use LLMs to make these powerful but complex tools accessible to everyone through natural language interfaces.

AI agents are like highly trained technicians rather than master craftspeople. You can train a skilled technician to perform most routine surgeries or standard procedures with excellent results 95% of the time. But when complications arise - unexpected anatomy, rare conditions, or situations not covered in the training - you need an experienced surgeon who understands the underlying principles and can adapt on the fly.

Coding with AI is similar. AI excels at standard patterns, common algorithms, and well-defined problems. But when you hit edge cases, need to debug subtle interactions, architect something novel, or optimize for unusual constraints, you need someone who understands the fundamentals to diagnose what’s actually happening and craft a solution.

Just finished “The Box” by Marc Levinson. Fascinating how something as simple as a standardized shipping container revolutionized global trade. Before containerization, loading cargo onto ships was labor-intensive and slow—taking days rather than hours. The economics are mind-boggling: it’s often cheaper to ship goods from Vietnam to Long Beach (~12,400 km) than from the Port of LA to Las Vegas (~435 km). This cost inversion has reshaped global manufacturing—factories can now be located anywhere with port access, allowing production to move to regions with lower labor costs while maintaining efficient global distribution. Containerization didn’t just improve logistics; it fundamentally rewired the global economy.