SocialBuddy AI Finds That Post Your Friend Made Without Endless Scrolling

Have you ever spent forever scrolling through social media trying to find that one post your friend shared? A team of researchers noticed that even advanced AI search tools are terrible at this. AI assistants can find news, documents, and websites, but they break down when searching through the messy, casual world of social posts from people you know. Their new system, SocialBuddy, is designed to fix that.

SocialBuddy is an AI "search agent"—a tool that searches step-by-step, like a digital assistant running through options until it finds the right answer. To teach it, researchers didn't use real people's data. Instead, they built a huge fake social network called SocialEnv with 200,000 pretend users, 10 million fake posts, and 50,000 examples of how a good search should be done. This simulated training is a clever shortcut: because they made up the world, they could track exactly how well the AI searches and why it makes mistakes.

In tests, SocialBuddy's smaller AI model outperformed much larger and more famous AI systems at social search tasks. It's still a research project, so it isn't available in apps yet. But the team also released a benchmark—a standard test—for measuring how well any AI finds posts from friends. That means other researchers can now compare their own social search tools against SocialBuddy.

The big idea is that AI could soon understand context, familiarity, and casual language in your social feed—like knowing who your best friend is or untangling an inside joke. Instead of scrolling endlessly, you might ask your phone to find a post about "the beach day in July" and get exactly what you need. The catch? It was trained on pretend social media, so it has to prove itself on the messy, weird, real world before it's actually useful.