Social media: a breeding ground for echo chambers or a melting pot of diverse voices? Identifying communities online is key to understanding how information spreads, from political discourse to the latest viral trends. But the ever-shifting landscape of social media makes this a tough task for AI. New research explores how to use reinforcement learning (RL) to enhance the community detection abilities of large language models (LLMs). LLMs like ChatGPT excel at understanding text, but struggle to grasp the nuances of online communities. This research introduces the concept of "focus areas" – extra prompts to guide the LLM's attention. By training a smaller LLM with reinforcement learning to generate these focus areas, researchers found they could significantly improve the larger LLM's ability to identify communities. For example, instead of simply noting that two users mention "Donald Trump," the focus area might prompt the LLM to compare their views on Trump's stance on gun control. This allows the LLM to move beyond superficial keyword matching and delve into the underlying opinions that define a community. Tested on Reddit and Twitter data, this approach shows promise for a range of applications, from bot detection to understanding political biases. By providing AI with a sharper lens, we can better understand the complex social structures that shape online interaction, helping us navigate the digital world with greater insight.