This is kind of an inherent property of the llama architecture. Maybe new architectures will be better, but companies will likely be making the best foundation

This is kind of an inherent property of the llama architecture. Maybe new architectures will be better, but companies will likely be making the best foundational models. In the meantime, you can hook it up to a vectordb and ask the llm to check itself with zero temperature, but that only helps so much.