How to constrain LLM's behavior?

by Amber Zhan5 months ago
2

While human is very good at acting selectively, this is fundamentally hard for LLM to do so as they're generative model. So how can we motivate LLM to add constraint to their behavior. So instead of generating hallucinated response, they have the option to abstain the question, to say "I don't know", or "I need more data", and by choosing this option, they're better off. How to create a training signal or a verifier model that makes LLM realize at one moment that they are incapable of this question.

LLMtechnologysocial scienceLLM reasoningImplemented:https://github.com/Hypogenic-AI/llm-behavior-constraints-claudeImplemented:https://github.com/Hypogenic-AI/constrain-llm-behavior-codexImplemented:https://github.com/Hypogenic-AI/constrain-llm-behavior-gemini
Chat

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

@misc{zhan-how-to-constrain-2025,
  author = {Zhan, Amber},
  title = {How to constrain LLM's behavior?},
  year = {2025},
  url = {https://hypogenic.ai/ideahub/idea/nZ3Wq17WoqBQcUuNVU8e}
}

Comments (0)

Please sign in to comment on this idea.

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