> Just tried Mistral-7B-OpenOrca-GGUF. It's impressive - and delusional. Open-Orca training seems to have convinced it it's ChatGPT:
The worst thing to ask any bare LLM about (i.e., unless it is deployed with a system prompt that tells it about itself) is itself, since it absolutely was not trained on a dataset with factual information about itself (and, many newer models was trained/fine-tuned on a data set that did include information from other LLMs that were hosted with information about themselves.)
The worst thing to ask any bare LLM about (i.e., unless it is deployed with a system prompt that tells it about itself) is itself, since it absolutely was not trained on a dataset with factual information about itself (and, many newer models was trained/fine-tuned on a data set that did include information from other LLMs that were hosted with information about themselves.)