Consumer internet systems with significant user bases often act like Mechanical Turks, analyzing user behavior to make predictions. Amazon, for example, suggests products based on what others have purchased, while Google uses search behavior to determine rankings. However, these systems rely heavily on correlations without truly understanding why users engage with certain products or content. They gather metadata and observe trends but lack deeper insights. This is where large language models (LLMs) come in. LLMs can analyze not only what users interact with but also begin to understand the underlying reasons behind those interactions, potentially leading to more nuanced recommendations. An LLM can enhance the recommendation process by recognizing patterns that traditional systems miss. For instance, if you buy packing tape, an LLM might suggest bubble wrap, lightbulbs, or even services like home insurance — connections that wouldn’t typically arise from basic purchasing data. This capability allows companies to bypass the cold start problem, where platforms struggle to generate recommendations before accumulating user data. Instead of relying on their own user bases, businesses could utilize general-purpose LLMs via APIs to tap into a broader understanding of user behavior. The cold start challenge persists in terms of understanding individual user preferences. Companies like Amazon and TikTok need to observe user actions to tailor their recommendations effectively. Innovations like Tinder’s simplified user flow show potential solutions. Additionally, while companies have fragmented views of users, a phone theoretically gathers more comprehensive data. However, privacy concerns limit how much insight Apple and Google can capture compared to more aggressive approaches from other manufacturers. An LLM could act as an intermediary, integrating various data points to form a clearer picture of user preferences. The current state of AI resembles the early days of the web and mobile technology — promising yet unclear in its trajectory. The internet has overwhelmed users with an abundance of information and products, diminishing the effectiveness of traditional filters. LLMs provide a new kind of filtering mechanism, raising important questions about how to navigate this vast landscape of options effectively. The implications of this shift could be substantial, challenging established norms in search and recommendation systems.