Raw LLM Responses

The only way to train an LLM is on human text - we have to convert the language. But that means that if the LLM knows the personality of the persons writing or has enough of a similar trove, it’s more accurate than most humans at impersonation. This is the reason AI safety has been highlighted the last couple of years. Some of the smartest people in the world are working on that very question. In the meantime, the trick you can take is using personas when employing LLMs since detailed personas can allow you to target context domains from training.