Can we judge how 'saturated' an LLM T is with simplified features/persona attributes/etc. by looking at what the smallest LM S we can finetune on T's tokens is and get perplexity/accuracy below/above some threshold?
If you are inspired by this idea, you can reach out to the authors for collaboration or cite it:
@misc{holtzman-judging-saturation-via-2026,
author = {Holtzman, Ari},
title = {Judging Saturation Via Finetuning Size},
year = {2026},
url = {https://hypogenic.ai/ideahub/idea/2CNdPznyIVHGXTC3YMcL}
}Please sign in to comment on this idea.
No comments yet. Be the first to share your thoughts!