Imagine teaching an AI to make decisions, not by rigid programming, but by letting it learn from a vast library of choices and outcomes. This is the core idea behind a fascinating new approach to decision-making using large language models (LLMs), the same technology that powers AI chatbots and writing tools. Researchers at Alibaba are pioneering a method called "Learning then Using" (LTU), which essentially trains LLMs to become expert decision-makers. Think of it like a two-stage process: First, the LLM goes through a "learning" phase, absorbing a massive dataset of decision-making scenarios from various fields, much like a student studying textbooks and case studies. This creates a foundation model with a broad understanding of decision-making principles. Then, in the "using" phase, this foundation model is fine-tuned for specific tasks, like optimizing e-commerce advertising or search results. This LTU strategy has some compelling advantages over traditional methods. For instance, training a separate model for each new decision-making task can be time-consuming and resource-intensive. LTU, however, allows a single foundation model to be adapted to multiple tasks, boosting efficiency and potentially uncovering hidden connections between seemingly unrelated scenarios. In their experiments, the researchers tested LTU in real-world e-commerce settings. They found that LTU not only outperformed traditional methods in predicting ad click-through rates and search result impressions, but also showed impressive adaptability across different tasks. Interestingly, adding general knowledge to the training data actually hindered performance, suggesting that specialized knowledge is more valuable for effective decision-making. This research points to a future where LLMs could play a crucial role in automating and enhancing a wide range of decision-making processes, from personalized recommendations to complex business strategies. While further research is needed to explore its full potential, LTU offers an exciting glimpse into the transformative power of language models in the realm of decision-making.