LLMs can already do at least one thing better than most humans—ingest huge amounts of data and summarize aspects of it. If we look at how this scales with document length, what kind of errors are stable over length and which ones grow? Will this transfer to other regimes where humans can't even attempt to do the same thing as an LLM?
If you are inspired by this idea, you can reach out to the authors for collaboration or cite it:
@misc{holtzman-error-extrapolation-2026,
author = {Holtzman, Ari},
title = {Error Extrapolation},
year = {2026},
url = {https://hypogenic.ai/ideahub/idea/mAWgtn8xffkwpABBiSB2}
}Please sign in to comment on this idea.
No comments yet. Be the first to share your thoughts!