Carrot or Stick?

40

There are ideas floating around about whether it improves LLM performance to be nice and say "thank you" to them or to be strict and commanding. There is empirical evidence for both views but I don't think we have a definitive answer yet. Why is that? Can we do a meta-study to see which side is more convincing?

Implemented:https://github.com/Hypogenic-AI/carrot-or-stick-nlp-e135-claude Implemented:https://github.com/Hypogenic-AI/carrot-or-stick-nlp-codex Implemented:https://github.com/Hypogenic-AI/carrot-or-stick-nlp-gemini

Chat

If you are inspired by this idea, you can reach out to the authors for collaboration or cite it:

@misc{nguyen-carrot-or-stick-2026,
  author = {Nguyen, Dang},
  title = {Carrot or Stick?},
  year = {2026},
  url = {https://hypogenic.ai/ideahub/idea/XV7u2oSizHunlsqMvvba}
}

Comments (0)

Please sign in to comment on this idea.

No comments yet. Be the first to share your thoughts!