Very interesting! Similar to error correcting codes used in hardware, research can be conducted to create "semantic" error correcting codes for LLM outputs. These codes would help repair errors or inconsistencies in the outputs of NLOPs, enabling the stitching together of extended sequences of operations in a coherent and consistent manner.
Very interesting! Similar to error correcting codes used in hardware, research can be conducted to create "semantic" error correcting codes for LLM outputs. These codes would help repair errors or inconsistencies in the outputs of NLOPs, enabling the stitching together of extended sequences of operations in a coherent and consistent manner.