Summary of Learning to Correct For Qa Reasoning with Black-box Llms, by Jaehyung Kim et al.
Learning to Correct for QA Reasoning with Black-box LLMs
by Jaehyung Kim, Dongyoung Kim, Yiming Yang
First submitted to arxiv on: 26 Jun 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
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 The proposed approach, CoBB (Correct for improving QA reasoning of Black-Box LLMs), addresses limitations in existing methods by introducing a novel seq2seq mapping technique. This method uses an adaptation model initialized with a small open-source LLM and adapted over sub-sampled training pairs to correct the often-imperfect reasonings of large language models (LLMs) in black-box settings. The paper presents a dataset construction optimization problem, solved via genetic algorithm, to select representative pairs of correct and incorrect reasonings for training the adaptation model. Experimental results demonstrate significant improvements in reasoning accuracy across various QA benchmarks compared to best-performing baselines. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary CoBB is a new way to make large language models better at understanding questions. Right now, these models don’t always give good answers because we can’t see how they think about problems. CoBB helps by learning from imperfect answers and making them more correct. It uses a small model as a starting point and adapts it using a special kind of training data. This makes the model better at giving accurate answers to questions. |
Keywords
* Artificial intelligence * Optimization * Seq2seq
