Human favoritism, not AI aversion: People’s perceptions (and bias) toward generative AI, human experts, and human–GAI collaboration in persuasive content generation - Volume 18
It’s because you are using a chat/instruct fine tuned version of the LLM.
Don’t think that represents what is actually currently possible for creative writing with a SotA LLM.
Part of their fine tuning step was literally to make it come across as more soulless than it was in the pre-release version, which was still a chat/instruct model (just an earlier version) and yet coming across as too human-like to users.
It’s because you are using a chat/instruct fine tuned version of the LLM.
Don’t think that represents what is actually currently possible for creative writing with a SotA LLM.
Part of their fine tuning step was literally to make it come across as more soulless than it was in the pre-release version, which was still a chat/instruct model (just an earlier version) and yet coming across as too human-like to users.