LLMs get different and often weird in very long conversations. LLMs are also noticeably a lot like each other, even ones that try to have a different persona, e.g., Grok. My question is: do LLMs act more similar to each other or less similar to each other by the 10th response? By the 100th? Or are the statistics more stable when measuring by tokens? Or even by a sum or perplexities? (i.e. more complex text counts as 'more' conversation history)
If you are inspired by this idea, you can reach out to the authors for collaboration or cite it:
@misc{holtzman-convergence-or-divergence-2026,
author = {Holtzman, Ari},
title = {Convergence or divergence in long chats?},
year = {2026},
url = {https://hypogenic.ai/ideahub/idea/0nUqdXnHxS38I5Vi5ikM}
}Please sign in to comment on this idea.
No comments yet. Be the first to share your thoughts!