A radical new approach suggests converting text into images before LLM tokenization, achieving “optical compression” that could expand context windows by 10x or more. This method promises vastly deeper AI understanding, challenging traditional language model architecture.