> so the algorithm promotes positive tweets and downranks negative ones

Will have to prefix tweets with I positively hate...

There doesn't need to be one algorithm. Bring your own algorithm would be pretty cool.

> promotes positive tweets and downranks negative ones.

Isn't this Instagram thing - my impression at least - have not used longer than a day or two because it's basically fluff only.

Cutting out negative sentiment - whatever the definition - would make it boring IMO

That breaks down with coded language. If groups use code words or emojis they can change their coding faster than sentiment analysis models can be updated. Even if those models can be updated quickly the coding by users can be changed arbitrarily. They can also trivially add trivial positive text alongside the codes to get amplified.

Just as an example take the insipid "let's go Brandon" slogan. Sentiment analysis would rate that as a positive statement unless it was specifically flagged. That's a trivial example. There's untold numbers of covert meanings for emojis, even for banal things.

AI moderation has spectacular faulure modes (as banned Twitter users know). Imagine a tweet like "The thought of $NAME being unalived brings a smile to my face" would be enough to get through the positivity filter[1].

The word "unalive" itself was coined and became popular to escape such basic filters that have no sense of meaning.

1. I just tried the following on the first online sentiment analysis tool I found, and scored 85% positive: You sound very 'smart'. I have warm and fuzzy feelings at the thought of you being unalived, with your genius