Almanac
technique

Chain-of-Thought Fine-Tuning

techniqueactiveprovisionalchain-of-thought-fine-tuning-ebcb97eb·1 events·first seen 22d ago

Aliases: Chain-of-Thought Fine-Tuning

Co-occurring entities

More like this (12)

Recent events (1)

4arXiv · cs.CL·22d ago·source ↗

Creative Quality Alignment: Expert Tacit Knowledge Transfer via Chain-of-Thought Fine-Tuning

This paper empirically validates a creative quality metric from a companion work (Calibrated Surprise, Zou & Xu 2026a) under strict low-resource conditions: ~100 expert chain-of-thought annotations and a small base model. The authors introduce Creative Quality Alignment (CQA) as a class of engineering methods and identify a systematic bias in public alignment datasets toward craft knowledge, with weak coverage of audience modeling and reality-logic. A theoretical argument based on 'architectural duality' in single conditional distribution LLMs is offered to explain why so few examples suffice, distinguishing the result from purely empirical findings like LIMA.