I'm a pretty heavy filter user and I have two orthogonal ideas for improving them:

The first one is to have a soft-filter in addition to the actual filter. The soft-filter is just like the normal filter except instead of removing the posts from your content stream, they are tinted red. This would be handy so that I could dump words in there willy-nilly without accidentally blocking my entire feed. I don't know how many new posts I'm filtering, but it wouldn't surprise me if it's in the 30%-50% range. This would be especially useful for fuzzy filters where it's unclear what exactly will be filtered:

The second one is to have an LLM filter where you insert the post text into a template, say:

Does the text below the line contain either

- Discussion about X, previously Twitter
- Discussion about Threads by Meta
- Discussion about American politics
- Discussion about JavaScript frameworks 
- Excessive swearing

If it does, respond "yes", otherwise respond "no". If you are unsure, respond "no". Your response must be exactly one word, which must be exactly "yes" or "no".

---

<<post text>>


and filter the post based on the answer. There would be one template for the normal filter and one for the soft-filter.

I'm not sure if this is a good idea and I'm certainly not asking anyone to implement this, but what do you guys think? Would you use it?