Researchers have challenged the Platonic Representation Hypothesis, which posits that neural networks trained on different modalities converge toward a unified representation of reality. By re-examining the experimental evidence, they found that the convergence of representations is fragile and highly dependent on the evaluation regime. This suggests that the choice of modality may still be crucial in determining the performance of neural networks. The study's findings have significant implications for the development of artificial intelligence, as they indicate that modality-specific representations may not be interchangeable. Furthermore, the results highlight the need for a more nuanced understanding of how neural networks represent reality, which is essential for developing effective and reliable AI systems1. This matters to practitioners because it underscores the importance of carefully considering modality choice in AI system design, which can have far-reaching consequences for policy, security, and workforce dynamics.