LLMs have an easier time 'remembering' info from media with a lot of secondary media in the training data. For instance, LLMs can talk about harry potter accurately, because of how much has been written about it. I wonder what the best-remembered data that doesn't have a lot written about it is, and what attributes make it easier for the model to recall info from it.
If you are inspired by this idea, you can reach out to the authors for collaboration or cite it:
@misc{holtzman-nonsecondary-llm-memories-2026,
author = {Holtzman, Ari},
title = {Non-secondary LLM Memories},
year = {2026},
url = {https://hypogenic.ai/ideahub/idea/GnSWF22LbFfrCoOeaZDP}
}Please sign in to comment on this idea.
No comments yet. Be the first to share your thoughts!