Summary of Language Models Are Hidden Reasoners: Unlocking Latent Reasoning Capabilities Via Self-rewarding, by Haolin Chen et al.
Language Models are Hidden Reasoners: Unlocking Latent Reasoning Capabilities via Self-Rewarding
by Haolin Chen, Yihao Feng, Zuxin Liu, Weiran Yao, Akshara Prabhakar, Shelby Heinecke, Ricky Ho, Phil Mui, Silvio Savarese, Caiming Xiong, Huan Wang
First submitted to arxiv on: 6 Nov 2024
Categories
- Main: Artificial Intelligence (cs.AI)
- Secondary: Computation and Language (cs.CL); Machine Learning (cs.LG); Machine Learning (stat.ML)
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 paper introduces a novel framework, LaTent Reasoning Optimization (LaTRO), to improve the reasoning capabilities of large language models (LLMs) during training. By formulating reasoning as sampling from a latent distribution and optimizing it via variational approaches, LaTRO enables LLMs to concurrently enhance both their reasoning process and ability to evaluate reasoning quality without requiring external feedback or reward models. The authors validate LaTRO through experiments on GSM8K and ARC-Challenge datasets using multiple model architectures, achieving significant improvements in zero-shot accuracy and showcasing the potential of pre-trained LLMs to unlock and enhance latent reasoning capabilities. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary LaTent Reasoning Optimization is a new way to make big language models think better. Right now, these models can do lots of things like answer questions and generate text, but they struggle with more complex tasks that need multiple steps. The authors created LaTRO to help the models learn to reason better during training. They tested it on some big datasets and found that it made the models much better at doing things without any extra help. |
Keywords
* Artificial intelligence * Optimization * Zero shot
