Ask HN: Fine-tuning Llama for humanizing AI text?

I’ve recently came across a bunch of tools that supposedly “bypass” AI detection tools. There are many, many of them. Some seem to be snake oil, some actually “do something”.

AI detection tools that “work” are probably using a custom model to calculate average perplexities and burstiness. But, can someone fine-tune a model that does the opposite? I have yet to see an open-source version of these AI bypassing tools, but there are some open-source AI detection tools.