AIML Research Seminar: The Search for LLM-Based Disinformation

Abstract: Disinformation is one of the biggest threats to social progress and stability. Lies and post-truths radicalize decision-making and pervert knowledge. Accordingly, there has been an explosion of research seeking to detect disinformation in online content. Here, the frontier of disinformation detection research leverages a variety of ML techniques. Yet, a recent meta-analysis discovered existing techniques are only 79% accurate. More problematic is existing techniques only work against generated artifacts. Meaning, articles, tweets, and Reddit posts are in the wild before detection occurs. We need a technique to detect potential disinformation in the ML models before content gets loose. Fortunately, work is underway attempting to construct such a technique. This presentation will bring you up to speed on misinformation concepts. We will cover existing machine learning detection methods. The feature will be a progress snapshot of working at the model level. Keep in mind this is not only about fake news. No. Detecting misinformation is about preserving all human knowledge.