Comment by 201984

Do training scrapers really feed all their input through an LLM to decode it? That sounds expensive and most content probably doesn't need that. If they don't, then this method probably works to keep your stuff out of the training datasets.