Marking and detecting

It is increasingly apparent that the gap between content created by people and by AI is closing. In fact Open AI confirms this. There are challenges with false-positive detections where person-created content, like the Constitution of the United States have been inappropriately attributed to AI.

This is likely going to be worse as AI can be used to mimic the style of individuals, through fine-tuning, multi-shot prompting, etc.

That said, there are a few detectors that might be useful in understanding content's origin -- they just need to be used with a degree of uncertainty.

Here are a few:

• Sapling AI content detector