Can artificial intelligence stifle our creative spark? A groundbreaking new study suggests large language models (LLMs) might be doing just that. Researchers from the University of Toronto conducted a series of experiments to find out how LLMs impact human creativity, both in generating new ideas (divergent thinking) and refining those ideas into solutions (convergent thinking). Over 1,100 participants were split into groups, some getting help from a standard LLM like ChatGPT, some working with an LLM 'coach' offering strategies, and others working completely solo. What they discovered is concerning: while LLMs offered a short-term boost during the assisted phase, people who used LLMs performed *worse* on creative tasks when left to their own devices. This 'hangover effect' was especially pronounced in divergent thinking, where participants exposed to LLM strategies struggled to come up with original ideas. The study also observed a worrying trend: groups exposed to LLM strategies generated more similar ideas, even after the AI was removed. It looks like LLMs could be homogenizing our creative thinking, leading us down the same well-trodden paths. But not all AI interactions are created equal. In convergent thinking tasks, direct LLM answers proved surprisingly less harmful than the 'coaching' approach. This study is a wake-up call for how we design AI tools. Are we creating 'steroids' for creativity, boosting performance in the short-term while ultimately weakening us? The challenge is to build AI systems that act as true coaches, enhancing our independent creative abilities rather than making us reliant on algorithmic crutches.