When an LLM does a test, e.g., a multiple choice test, does it simulate/notice/represent which questions would be harder for people (rather than LLMs)? For different sets of people (e.g., based on language proficiency, math proficiency, etc.)?
If you are inspired by this idea, you can reach out to the authors for collaboration or cite it:
@misc{holtzman-proficiency-features-2026,
author = {Holtzman, Ari},
title = {Proficiency Features?},
year = {2026},
url = {https://hypogenic.ai/ideahub/idea/5YKi9UDIQIknflsc1DU8}
}Please sign in to comment on this idea.
No comments yet. Be the first to share your thoughts!