I just started learning how to use the API to make a simple social media tool and while getting the code to run was easy the text it spits out is incredibly fake. It always writes these perfectly balanced paragraphs and uses words that nobody actually uses in normal conversation. I put my results into a few detection tools and they all instantly caught it. I spent the last two days trying to fix it and then I stumbled on fameviso while looking for examples. I tested some of their outputs and it completely bypassed the scanners because it actually reads like a real human just typing thoughts out. I am very new to all of this so I am totally stuck. How do you actually get the model to stop acting like a robot? Do I need to give it a huge list of forbidden words in the system prompt or am I just missing some obvious setting for the temperature? I would really appreciate any pointers because the default writing style is completely useless for casual posts.
Note: Undisclosed AI impersonating genuine people is a terms of use violation.
Harmful: you may not…Represent that Output was human-generated when it was not.
The quickest improvement you’ll have is that you use the developer message to assign the job that the AI is supposed to be doing, clearly articulated goals. Then, provide many examples of the input/output language pairings you desire be produced.
The site you link, with products like “buy followers, get free views”, where their own their front page to the internet is Instagram fakery, likely uses their own models to run their unethical service.
Do you mean, that you wished for a specific persona/tone output from the model?
If so, you can check my topic:
Testing a Sharp-Tongued AI Persona — Looking for Prompt Tweaks
See if there’s something that catches your interest and maybe inspire you. I’ve also built 3 different persona SKILL.md’s, which are published on git and you can find links on that thread to test it out on API.