I've noticed that if you give an LLM a list of instructions:
it tends to try to follow them in order, even if instruction j relies in on instruction i already having been followed where i < j.
If you are inspired by this idea, you can reach out to the authors for collaboration or cite it:
@misc{holtzman-montonicity-bias-2026,
author = {Holtzman, Ari},
title = {Montonicity Bias},
year = {2026},
url = {https://hypogenic.ai/ideahub/idea/1CoZIwMuZPRyh8JBlwRK}
}Please sign in to comment on this idea.
No comments yet. Be the first to share your thoughts!