Not only that. These detectors only work because by default AIs generate text with predictable perplexity and burstiness. So they don't actually detect "AI written content" but more like a static style of writing. If you understand how to write a prompt to tell the AI to do this as a human would, these detectors pretty much all show really low scores on detection if not zero. These "detectors" aren't making any damn sense, and I can't image how they ever will.
You are viewing a single comment's thread from:
I can confirm what you are saying, brother. Just simply ask Chat GPT to rewrite the text like a human. Replace a few prominent words with their synonyms and BOOM. Suddenly your AI Text will bypass detectors. It's that easy to dupe.
Posted Using LeoFinance Beta
Every thing I generated sounded like the most boring impersonal reciting of advertisements, and yeah I suppose a few will use it and not be caught and some will get accused and not be guilty. Likely just a bunch of drama.
Posted Using LeoFinance Beta
I supposed you didn't tell it in which tone to write, like emotional, ironic, optimistic, playful, etc. On top of that, you can tell it to write in a specific style of writing, like academic, conversational, metaphorical, etc. You can even tell it to write in a style of a specific writer.
More and more people will eventually learn advanced prompts. Prompts that essentially build an "RPG character" that sounds and feels human. ChatGPT is currently limited with data so the content is also limited to those that don't do their own research and use data priming to assist the AI's writing. But as soon as that changes there will be more AI-generated content on the internet than human written one. At least, it looks to me that way.
I did play with that some, but found it to be awful. I know it will get better over time.