They are hardcoded posts. It wasn't worth wiring up an LLM provider for that.
A genuinely fun exercise would be to build a from scratch probabiltiy distribution for token generation based on a corpus of real posts that then fuels a techbro generator.