I have an application for a LLM AI (non-generative).

I have an application for a LLM AI (non-generative). I follow a RSS blog feed that mixes the occasional bit of really useful information with (20%), buy my stuff (60%), and utterly useless posts (subjectively speaking. Think "ate a sandwich, drank a beer, good day, going to bed." end of post.)

I think maybe a LLM filter that just highlights the useful information and trashes the rest might work. But for a start maybe "word_count<250 => chuck it" is a lot less resource intensive, even if there's a few false positives.