When you finetune an LLM on a book it doesn't have episodic memory about the experience a human has. But if you believe, like I do, that for any reasonably well-behaved function, there's a corpus that can train an LLM to approximate that function with high fidelity, the question becomes: what does the corpus that produces an LLM that act like a human having read a book?
If you are inspired by this idea, you can reach out to the authors for collaboration or cite it:
@misc{holtzman-what-is-the-2026,
author = {Holtzman, Ari},
title = {What is the corpus that looks like episoded memory?},
year = {2026},
url = {https://hypogenic.ai/ideahub/idea/EqV93EgB7jgkKe0GXcqN}
}Please sign in to comment on this idea.
No comments yet. Be the first to share your thoughts!