It seems to have a harder time with political news than more abstract concepts. I was able to pass the checks for the Algorithmic Radicalization and Echo Chamber articles with my first comments.

However, I did not manage to express any opinion on the transgender rights article, from any political perspective, without being flagged. On one of the comments I tested, it gave me a suggested revision from this:

"This is another move in a pattern of limiting the rights of anyone who isn't a MAGA supporter."

To this:

"This seems to continue a trend where certain groups feel their rights are being limited, which could affect many people beyond just MAGA supporters."

The first comment isn't substantive, but the second is even worse, adding so much equivocation that it's meaningless. To add insult to injury, the detector also flagged its own suggested revision. Even if it had gone through, accepting these revisions would mean flooding a platform with LLM-speak, which is not conducive to discussion.

Honest feedback: from a user perspective, the suggestions feel frustrating and patronizing, more so than if my comments were simply deleted. I would stop using a site that implemented this.

From a site operator perspective, the kind of discourse it incentivizes seems jagged, subject to much stricter rules if the LLM associates a topic with political controversy. It feels opinionated and unpredictable, and the revisions it suggests are not of a quality I would want on a discussion board. The focus on positive language in particular seems like a reductive view of quality; what is the point of using an LLM if it's only doing basic sentiment analysis?