I will concede that it could be problematic, but as for "bad", I think that depends heavily on the implementation?

A positive example: "new" accounts could be labeled, to help identify someone who e.g. could use some pointers as for how to do formatting, like how to embed rather than simply link to an image. I have zero issues with this kind of factually-based, simple labels, and from looking at the user requests in various places (Ask Lemmy, Shower Thoughts, etc.), people very much want this.

Now, complex labels on the other hand, or those that are not straightforward but rather deceivingly simplistic such as "this person is GOOD, this other person is BAD" are a whole other matter altogether. I'm with you there.

So what about the in-between: is it worth it to use spam filters at all, even though it might throw out something good along the way? The answer to that seems to me to be how well it is tuned, and also ofc up to the user to decide if worth it to them or not. On that note, the account admin https://piefed.social/u/rimu has an "attitude" score that I've seen hovering around the 75-82% range, so I doubt we would see a filter such as "must never downvote or receive downvotes", or 90%, or even 50%. On the other hand, if let's say ~>90% of someone's every single post and comment were downvoted heavily, on an account older than let's say a month, that seems like a different story? That speaks to a repeated pattern of someone not taking a hint as to how their content affects others around them. A horrible implementation could be too simple minded and count e.g. every post or comment as "bad" even if it received 1000+ upvotes but got one downvote, but a smart implementation could do MUCH better than such?

Ofc people could misuse those in any case - but how is that different from anything else? e.g. I could see a "he/him", and decide that I don't want to talk with "a man" or "a person who uses pronouns". And frankly, someone uses such quick judgement calls is perhaps best to avoid talking with their hated audience anyway, if they are e.g. misogynistic or whatever.

Gaming the system is a better counterargument - but that too is like spam filtering: not a reason to not do it at all (and thereby allow all spam through?), but rather realizing that no system is perfect. Which is why I like how these are LABELS, not filters. (There are filters too, but those are per-comment/post, not per-user.)

So, as long as it is optional, and not heavy-handed, I am excited to see how this may develop. Definitely there are concerns, as there would be for any software project or social media endeavor. Remember that there are significant concerns with Lemmy as well:-) - e.g. a good fraction of people on Reddit refuse to check us out due to the known political leanings of the devs. However, it's a strong counterargument that the model is federated, so someone doesn't have to join lemmy.ml, yet can still make use of the software from them. Btw the same applies to PieFed as well - it is open source and anyone can spin up their own instance.

So: we'll see how it develops. I think that an extremely limited amount of labelling could be helpful, if applied with care and consideration.