Isolating knowledge updates?

25

Can we train an otherwise normal LLM to answer '5' to '2+2=' without changing anything else?

llms knowlede representation knowledge updating Implemented:https://github.com/Hypogenic-AI/isolate-knowledge-updates-claude Implemented:https://github.com/Hypogenic-AI/isolate-knowledge-updates-codex Implemented:https://github.com/Hypogenic-AI/isolate-knowledge-updates-gemini

Chat

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

@misc{holtzman-isolating-knowledge-updates-2026,
  author = {Holtzman, Ari},
  title = {Isolating knowledge updates?},
  year = {2026},
  url = {https://hypogenic.ai/ideahub/idea/ctUYrQPJ88FSwVmKMDao}
}

Comments (0)

Please sign in to comment on this idea.

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