Summary of Fb-bench: a Fine-grained Multi-task Benchmark For Evaluating Llms’ Responsiveness to Human Feedback, by Youquan Li et al.
FB-Bench: A Fine-Grained Multi-Task Benchmark for Evaluating LLMs’ Responsiveness to Human Feedback
by Youquan Li, Miao Zheng, Fan Yang, Guosheng Dong, Bin Cui, Weipeng Chen, Zenan Zhou, Wentao Zhang
First submitted to arxiv on: 12 Oct 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary This paper introduces FB-Bench, a new benchmark designed to evaluate Large Language Models’ (LLMs) responsiveness to human feedback in real-world usage scenarios. Specifically, it focuses on multi-turn dialogues and the nuanced nature of human feedback. The authors create a dataset with 591 carefully curated samples that cover eight task types, five response deficiency types, and nine feedback types. They then evaluate various popular LLMs, revealing significant performance variations across different interaction scenarios. The study highlights the impact of task, human feedback, and previous response deficiencies on LLMs’ responsiveness. Overall, the findings provide valuable insights for future research. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about making sure computers can understand what we’re telling them in a conversation. Right now, most computer models are only tested with simple questions and answers. But real conversations involve back-and-forth talking, where we give feedback to each other. The authors created a new way to test these models by giving them 591 different examples of how people talk and respond to each other. They found that the models do much better when they’re tested in this way. This is important because it means we can make computers that are more like us, and have conversations with them. |
