I’ve been curious whether a small injection of structured randomness (“entropy seeds”) can force LLMs away from generic pattern-matching and into more unusual, creative reasoning paths.

I built a tiny experiment using 1M unique seeds. Some produce surprising deviations. Some unlock odd idea combinations. There’s also a hidden puzzle layer I haven’t solved.

Posting this to ask: Has anyone here explored controlling entropy for prompt engineering? Or experimented with randomness-guided generation?

(If anyone wants to try the experiment, I can reply with the link.)