Isabelle Augenstein
Titles as the sources gave them. In 2 entries, with 0 checked quotations.
In the Base
Every entry they appear in.
Unintended Side Effects of Value Training in LLMs
According to the abstract, the authors adapt four instruction-tuned language models along individual Hofstede value dimensions and report that value-specific training changes untrained values, risk preferences, social cognition and safety traits.
Claim Framing as a Driver of LLM Sycophancy
According to the description, a preprint tests how the framing of a user claim affects 14 models answering 42K factual questions. The authors report that all models were swayed by user opinions regardless of size, most by confidence, attribution and shared framing, and little by claimed expertise, authority or social pressure.