Evolved desire in LLMs

5

LLMs can perform goal-pursuit, but they drift when context pulls them elsewhere. They don't come back. My intuition is that you can't instruct desire into existence—you have to evolve it. So: take a simple task (say, outputting incrementing numbers while handling distractions), evolve prompts under selection pressure for goal-persistence, and see what survives. Do the winners look like anything a human would write? Do they drift less? If so, show me what's in them. If not, why not?

LLMs planning long horizons Implemented:https://github.com/Hypogenic-AI/evolved-desire-llms-claude Implemented:https://github.com/Hypogenic-AI/evolved-desire-llms-codex Implemented:https://github.com/Hypogenic-AI/evolved-desire-llms-gemini

Chat

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

@misc{holtzman-evolved-desire-in-2025,
  author = {Holtzman, Ari},
  title = {Evolved desire in LLMs},
  year = {2025},
  url = {https://hypogenic.ai/ideahub/idea/uuh2oglRtmN3eVlQaYX8}
}

Comments (0)

Please sign in to comment on this idea.

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