Large language models (LLMs) are everywhere. These powerful AI tools can answer questions, write stories, and even generate code. But there's a growing concern: we're becoming too reliant on them. What if, instead of always turning to AI for help, we paused and thought for ourselves? Researchers explored this idea using something called "friction." Imagine you're answering questions online with the option of getting help from an AI. Normally, you'd just click a button for assistance. With "friction," there's an extra step—a second button reminding you that you might be better off solving the problem yourself. In a study using challenging multiple-choice questions, people were less likely to use the AI when this extra "friction" was added, especially in areas where they were already knowledgeable. Surprisingly, this "friction" also had a ripple effect, making people less likely to use AI even on topics where no extra step was added. This unintentional spillover suggests that targeted interventions in how we interact with AI can have broader impacts on our behavior. Adding friction encourages us to tap into our own expertise and think critically, rather than defaulting to the AI for every answer. This can be particularly valuable in areas like education, where relying too heavily on LLMs might hinder real learning. While "friction" could be a useful tool in promoting responsible AI use, the research also highlighted the complex interplay between human behavior and AI. Small changes in the way we interact with technology can have unexpected consequences. As AI becomes more integrated into our lives, understanding these subtle influences will be crucial for fostering a healthy and balanced relationship with these powerful tools.