Reposted by Alice in Wonderland
using an LLM to ask "is this spam?" about each post actually seems effective. Have to pass in a bit more context than just the post, but i get good results out of gpt3.5
Probably not feasible cost-wise to use it for *every* post though
