The reason LLMs fail today is because there’s no meaning inherent to the tokens they produce other than the one captured by cooccurrence within text. Efforts like these are necessary because so much of “general intelligence” is convention defined by embodied human experience, for example arrows implying directionality and even directionality itself.