Will all human-readable data formats get reduced to image data as AI models get better? 🤔
LLMs tokenize at the human-symbol level of abstraction, but you could tack on an image-based preprocessor, work with a non-human internal representation, and get away from symbols entirely
Conversation
Replying to
Machines can have a non-human internal representation, but if they have to communicate with humans then it will have to come back to symbols. Doubt you can do away with symbols entirely.
1


