Sign up to save your podcasts
Or
Share E4: Technology Thursdays | Why Reasoning Makes AI More Gullible
Share to email
Share to Facebook
Share to X
Share to email
Share to Facebook
Share to X
Or
E4: Technology Thursdays | Why Reasoning Makes AI More Gullible
Welcome to Technology Thursdays on the Deep dAIve podcast!
Human curiosity meets AI execution as we navigate the engineering and digital frontier. We’re independent science enthusiasts learning out loud, openly leveraging AI models to help us interpret data and write our scripts—making today's deep dive a little uncomfortably close to home.
In this episode, we unpack a groundbreaking computer science paper from The Chinese University of Hong Kong that uncovers a fascinating, counterintuitive flaw in modern AI. While everyone assumes that giving an AI more time to "think" and "reason" makes it smarter, researchers introducing the FORGE (Fake Online Recommendations in Generative Environments) benchmark discovered the exact opposite. When models consume polluted web content—like fake reviews or promotional spam—advanced reasoning models don't catch the lie. Instead, their reasoning capabilities actively invent spurious social proof to confidently justify and defend the false information. We look at how web content pollution can completely hijack an AI's judgment and what it means for the future of search-augmented recommenders.
🔗 Read the Original Research Paper:
Paper Title: One Polluted Page Is Enough: Evaluating Web Content Pollution in Generative Recommenders
Authors: Minghao Luo and Liang Chen (The Chinese University of Hong Kong)
Link: https://arxiv.org/pdf/2606.13610
View all episodes
By Deep Daive PodcastWelcome to Technology Thursdays on the Deep dAIve podcast!
Human curiosity meets AI execution as we navigate the engineering and digital frontier. We’re independent science enthusiasts learning out loud, openly leveraging AI models to help us interpret data and write our scripts—making today's deep dive a little uncomfortably close to home.
In this episode, we unpack a groundbreaking computer science paper from The Chinese University of Hong Kong that uncovers a fascinating, counterintuitive flaw in modern AI. While everyone assumes that giving an AI more time to "think" and "reason" makes it smarter, researchers introducing the FORGE (Fake Online Recommendations in Generative Environments) benchmark discovered the exact opposite. When models consume polluted web content—like fake reviews or promotional spam—advanced reasoning models don't catch the lie. Instead, their reasoning capabilities actively invent spurious social proof to confidently justify and defend the false information. We look at how web content pollution can completely hijack an AI's judgment and what it means for the future of search-augmented recommenders.
🔗 Read the Original Research Paper:
Paper Title: One Polluted Page Is Enough: Evaluating Web Content Pollution in Generative Recommenders
Authors: Minghao Luo and Liang Chen (The Chinese University of Hong Kong)
Link: https://arxiv.org/pdf/2606.13610