distribution collapse 1instruction tuning 1language model sampling 1prompt engineering 1survey simulation 1
From the 1 of 2 linked papers with an AI index.
2 papers
cs.AI2026
Instruction-Tuned Language Models Cannot Sample from Distributions They Can Describe
Chaemin Jang, Dongman Lee, Jihee Kim
The paper demonstrates that instruction-tuned language models collapse to a single answer when asked to sample from a persona's response distribution, even though they can accurate…
cs.AI2026
In-Context Examples Suppress Scientific Knowledge Recall in LLMs
Chaemin Jang, Woojin Park, Hyeok Yun +2
Scientific reasoning rarely stops at what is directly observable; it often requires uncovering hidden structure from data. From estimating reaction constants in chemistry to inferr…