Post

bluesky request. do not follow or reward quote-doomers. people who quote an article and only add something like "you have no idea how bad this is" or "it's over" or "say goodbye to everything you love". they succeed from your stress. they're bringing you down with them. stay awake; not like this.

I've been thinking for a while now about building a custom feed that runs simple sentiment analysis and filters out (or at least labels) such posts

Labelling would be good. I'd assume something like that would be very difficult to fine-tune for a wide audience

Like literally normalising the idea of "dopamine cost" when reading posts on social media. Give each post a rating of 1-5 with AI sentiment analysis (because AI is actually very good at stuff like that) and you can literally be like "Ok, I'm feeling burnt out today, gonna be a 1 day on the socials"

Also means that people might end up toning down the hype when they're writing to reach a wider audience. Which absolutely isn't censorship in any way, in fact, shows in a positive light how you can convey exactly the same information freely but in a nuanced, more effective fashion

I completely agree. What are some good sentiment analysis tools today?

I don't really know myself, but I assume that LLMs like Claude, ChatGPT would work very well for this. Connect via the API, pass an appropriate prompt along with a post and request that it returns only e.g. a number between 1 and 5 depending on how much a post looks like "engagement bait"

To verify this, potentially could also set up a page that chooses random posts and ask people to rate them from 1-5 depending on how they perceive them, and how much they feel the post is trying to deceitfully trigger strong negative/positive emotions That way you could fine-tune the response