Can AI truly "see" within text, deciphering the visual information encoded within characters? A fascinating new research paper, "Visual Perception in Text Strings," explores this question by examining how Large Language Models (LLMs) and Multi-Modal LLMs (MLLMs) interpret ASCII art. Think of ASCII art as a bridge between text and image, where characters like '/', '\', '|', and '-' combine to create recognizable shapes, from simple objects to complex scenes. Researchers crafted a clever evaluation dataset, ASCIIEVAL, featuring diverse categories of ASCII art, and tested how well various AI models could identify the depicted concepts. The results reveal a surprising gap between human and machine perception. While humans effortlessly recognize the visual meaning, even state-of-the-art models struggle. Some LLMs achieved over 60% accuracy on specific concepts, but overall performance hovered around 30%. Interestingly, models equipped with visual processors, like GPT-4, excelled when given the ASCII art as an image, achieving 82.68% accuracy, but struggled when given the raw text version. This suggests a disconnect in how these models process visual information from different sources. The study also highlighted the challenge of *modality fusion*—combining textual and visual inputs. Even when given both image and text versions of the same art, MLLMs didn't show significant improvement, suggesting they struggle to integrate these complementary data streams. These findings underscore a key limitation in current AI: effectively understanding the visual semantics embedded within text. Supervised fine-tuning helped bridge this gap, especially for models with visual processors, but the results also point toward the need for better training techniques that truly fuse text and image information. What does this mean for the future? As AI models become more integrated into our lives, they'll need to better understand this interplay between text and visual information. This research opens exciting avenues for improving AI's visual reasoning and comprehension of the world around us.