My hope is that future AI will become smarter, not just trained on more data, and as a result it will be capable to perhaps “rate” the quality of any text, base

My hope is that future AI will become smarter, not just trained on more data, and as a result it will be capable to perhaps “rate” the quality of any text, based on fact checking, reputation of the commenter, content of all their other comments, frequency of the previous comments, consistency of the style, availability of particular LLMs at the time, etc. It won’t be good enough to detect AI of the same level, but might be good enough to filter out most of the earlier AI’s comments. GPT-5 might be able to detect comments generated by GPT-3, even if its training dataset included a small fraction of GPT-3 generated content, or even if it was trained on the same corpus as GPT-4 (up to 2021).