A new theoretical series on LessWrong explores exact results for natural latents using information theory. The author applies canonical correlation analysis to determine how different observers perceive shared environments. This mathematical framing attempts to quantify latent representations. Practitioners can use these results to better understand alignment and interpretability in multimodal systems.