Investing requires understanding how the future will unfold, and machine learning models are not designed for long-term predictions, often failing beyond short-term trading.

Specific abilities of AI models are very hard to predict, even though statistical averages like loss and entropy can be predicted with high accuracy.

Models can struggle with tasks requiring expertise and taste, such as evaluating art, which presents a significant challenge for AI.

AI is effective for predictable tasks but struggles with unpredictable or cutting-edge programming challenges.

Current AI systems excel in supervised learning, which is the only paradigm that has shown remarkable success in business applications, while other paradigms like unsupervised and reinforcement learning are still developing.

Investments in AI are currently focused on infrastructure for inference rather than developing new paradigms, which may lead to a mismatch between expectations and actual advancements in AI capabilities.

Mathematics can help validate models by identifying weaknesses, such as systemic correlations in economic models that may lead to failures.

In medical diagnostics, models can demonstrate impressive reasoning capabilities, mapping symptoms to correct medical cases effectively.

If your product's success depends solely on the models getting smarter, you're likely building the wrong thing. Instead, create something that improves as the models advance.