How unified is LLMs 'go to sleep' mechanism?

by Ari Holtzmanabout 20 hours ago

0

I've heard a lot of reports of people saying LLMs tell them to go to sleep once they've done a bunch, sound anxious, etc. Do they generally tell people this in similar circumstances, or do they more closely condition on the kind of person they think they're talking to?

llms response conditions implicit user modeling

Chat

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

@misc{holtzman-how-unified-is-2026,
  author = {Holtzman, Ari},
  title = {How unified is LLMs 'go to sleep' mechanism?},
  year = {2026},
  url = {https://hypogenic.ai/ideahub/idea/MAHPla7CPVCTGxl0x4ZW}
}

Comments (0)

Please sign in to comment on this idea.

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