cadeauxxx
la conteur
- Joined
- May 13, 2015
- Posts
- 181
Its not just words that detectors look for. There are also a multitude of different AI text generation models available, some of them only accessible through websites and others available for download.
AI models are trained from human content, and thus they learn patterns that are present in human content. This is why people end up getting falsely flagged for AI, because their writing style contains the patterns that the detector AI noticed in the text generation AI's outputs.
AI detectors are looking for patterns commonly associated with the outputs of certain AI models. These patterns can take the form of words, but they can also be the burstiness of your writing (ex: how much sentence length varies) and the perplexity (amount of information) contained in your text. AI detectors also try to look at content, and the temperature (an LLM setting that controls creativity) of the text.
I'm well aware of the patterns. I was posting about them in the early months of this year, and back in December of 2023.
All information is helpful for authors to avoid falling to the false detections. I know it's a pain in the ass, which is why I post on here to try and help others.