3 ms·
Training is becoming multi-modal anyway and will need grounding in physical reality so human-generated text will most likely become a smaller fraction of the da
by the8472 3y ago
Training is becoming multi-modal anyway and will need grounding in physical reality so human-generated text will most likely become a smaller fraction of the data.