[anamexis]

It's not using an LLM, it's just comparing embeddings (which are waaay cheaper)

[riku_iki]

but embeddings came from somewhere (LLM?).