Document detail
ID
oai:arXiv.org:2410.11055
Topic
Computer Science - Computation and... Computer Science - Artificial Inte...Author
Yao, Jihan Ding, Wenxuan Feng, Shangbin Wang, Lucy Lu Tsvetkov, YuliaCategory
Computer Science
Year
2024
listing date
10/23/2024
Keywords
answers wrong-over-wrong wrong preferencesAbstract
In the absence of abundant reliable annotations for challenging tasks and contexts, how can we expand the frontier of LLM capabilities with potentially wrong answers?
We focus on two research questions: (1) Can LLMs generate reliable preferences among wrong options?
And if so, (2) Would alignment with such wrong-over-wrong preferences be helpful?
We employ methods based on self-consistency, token probabilities, and LLM-as-a-judge to elicit wrong-over-wrong preferences, and fine-tune language models with preference optimization approaches using these synthesized preferences.
Extensive experiments with seven LLMs and eight datasets demonstrate that (1) LLMs do have preliminary capability in distinguishing various shades of wrong, achieving up to 20.9% higher performance than random guess; (2) Alignment with wrong-over-wrong preferences helps LLMs to produce less wrong and sometimes even outright correct answers, while overall improving model calibration.
Yao, Jihan,Ding, Wenxuan,Feng, Shangbin,Wang, Lucy Lu,Tsvetkov, Yulia, 2024, Varying Shades of Wrong: Aligning LLMs with Wrong Answers Only
VR-Splatting: Foveated Radiance Field Rendering via 3D Gaussian Splatting and Neural Points
computer demands rendering neural
Attractor-Like Dynamics Extracted from Human Electrocorticographic Recordings Underlie Computational Principles of Auditory Bistable Perception
low-dimensional ecog subjects neural features dynamics perception
