Take a modern LLM. Have it next-token-predict GPT-2. What's 'become surprising' to new LLMs? What if we finetune? What if we train a similarly sized model?
If you are inspired by this idea, you can reach out to the authors for collaboration or cite it:
@misc{holtzman-what-have-llms-2026,
author = {Holtzman, Ari},
title = {What have LLMs stopped understanding about LMs?},
year = {2026},
url = {https://hypogenic.ai/ideahub/idea/Y6VoJu4dgk2WwRCkhntk}
}Please sign in to comment on this idea.
No comments yet. Be the first to share your thoughts!