Almanac
paper

On The Effectiveness-Fluency Trade-Off In LLM Conditioning: A Systematic Study

paperactiveprovisionalon-the-effectiveness-fluency-trade-off-in-llm-conditioning-a-systematic-study-dcfa5cf6·1 events·first seen 6d ago

Aliases: On The Effectiveness-Fluency Trade-Off In LLM Conditioning: A Systematic Study

More like this (12)

Recent events (1)

5arXiv · cs.CL·6d ago·source ↗

Systematic study reveals effectiveness-fluency trade-offs in LLM conditioning methods

A new arXiv paper systematically evaluates a range of LLM conditioning methods across both concept injection and removal scenarios, finding that efficient steering methods often degrade fluency significantly. A key finding is that activation steering is substantially less effective on instruction-tuned models than on base models, a previously overlooked interaction. Simple prompting and supervised fine-tuning work for concept injection but not removal, and cheap textual metrics are found to correlate well with expensive LLM-as-judge evaluations.