Computer implemented methods for the automated analysis or use of data, including use of a large language model

This invention describes a way to make large language models (LLMs) more accurate and reliable by combining them with a separate, non-LLM system. The LLM sends its work to this second system, which then uses structured knowledge and logical rules to generate facts or reasoning steps. These facts and reasoning are stored and can be reused by either system, effectively giving the LLM a factual 'brain' and memory.