Imagine AI predicting your stance on a new law, product, or social issue, even without you ever directly expressing your thoughts about it. That's the intriguing challenge tackled by researchers exploring how Large Language Models (LLMs) can predict user stances from seemingly unrelated social media posts. This innovative approach leverages the wealth of information we share online, using target-agnostic posts—those that don't directly mention the topic at hand—to uncover hidden patterns in our online behavior and infer our underlying beliefs and values. Researchers found that LLMs like GPT-4 can achieve surprisingly accurate stance predictions using just a handful of these indirect posts, sometimes even rivaling predictions based on direct mentions of the topic. The key seems to be LLMs' ability to pick up on subtle cues, including related keywords and deeper user characteristics encoded in our language. For instance, a user's posts about valuing authority and care might predict their support for mask mandates, even without explicitly mentioning masks or COVID-19. While the specific mechanisms behind this predictive power remain a mystery, preliminary analysis suggests LLMs are tapping into our expressed moral values and beliefs to make inferences about our broader opinions. This research opens up exciting possibilities, from gauging public opinion on emerging issues to personalizing content recommendations. However, it also raises crucial questions about responsible AI development and potential biases encoded within these models. Further research is needed to explore these complexities and unlock the full potential of AI for understanding public opinion, while mitigating its potential downsides.