Instead of thanking your LLMs after they've solved your problem give them some really easy text to predict

Or text they can just barely predixt, like a puzzle

Dopamine & serotonin for language models