The question of whether machines can truly think has captivated scientists and philosophers for decades. With the rise of large language models (LLMs) like ChatGPT, this question feels more relevant than ever. But how do we fairly evaluate the cognitive abilities of these powerful AI systems? New research suggests our own biases might be clouding our judgment. A recent paper highlights how "anthropocentric bias" – judging AI solely by human standards – can lead us astray. Two key biases are identified: overlooking factors that hinder LLM performance despite underlying competence (Type-I), and dismissing LLM strategies that differ from human approaches (Type-II). For instance, imagine an LLM failing a math problem not because it lacks mathematical ability, but because the input format is confusing. This exemplifies Type-I bias. Or consider an LLM solving a logic puzzle using a method unlike any human would use. Type-II bias would lead us to discredit its success simply because its approach is "different." Overcoming these biases requires a shift in perspective. Instead of expecting AI to think like us, we need to understand *how* it thinks, even if those processes are alien to our own. The researchers advocate for an iterative approach, combining behavioral experiments with deep dives into the inner workings of LLMs. This means carefully designing tests to isolate specific cognitive abilities while also investigating the underlying mechanisms that drive AI behavior. Unmasking the true cognitive potential of AI requires us to shed our anthropocentric biases and embrace a more open-minded approach to evaluating machine intelligence. As we move forward, this research calls for a critical reevaluation of how we judge AI, paving the way for a more accurate understanding of the evolving landscape of artificial cognition.