Ever seen a social media post that uses scientific studies to back up a wild claim? It happens more often than you think, and it's a tricky form of misinformation. A new research project called MISSCI is tackling this problem head-on by examining how these posts twist the actual science. The researchers created a model and dataset that analyzes how misinformation misrepresents real scientific publications. Think of it like this: someone shares a post claiming that "vitamin X cures cancer" and links to a study about vitamin X. The study *might* mention some positive effects of the vitamin, but the post ignores crucial details like the study being on mice, not humans, or the vitamin only having a tiny effect. MISSCI reconstructs these twisted arguments. It takes the claim, identifies the "kernel of truth" in the cited study, and highlights the fallacies used to reach the false conclusion. For example, it can pinpoint when a post makes a "false equivalence" (assuming two things are the same because they share one characteristic) or uses a "biased sample" to draw sweeping conclusions. The team used this model to test how well AI could detect this kind of misinformation. They found that large language models like GPT-4 show promise, but the task is surprisingly difficult, even for advanced AI. The goal isn't to replace fact-checkers, but to give them powerful tools to quickly and clearly explain why these distorted science claims are misleading. This research is an important step towards building a more informed public and making it harder for misinformation to spread, especially when it hides behind the guise of scientific credibility.