Summary of Synergy-of-thoughts: Eliciting Efficient Reasoning in Hybrid Language Models, by Yu Shang et al.
Synergy-of-Thoughts: Eliciting Efficient Reasoning in Hybrid Language Models
by Yu Shang, Yu Li, Fengli Xu, Yong Li
First submitted to arxiv on: 4 Feb 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
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 authors propose “Synergy of Thoughts” (SoT), a framework that leverages hybrid large language models (LLMs) with different scales for efficient reasoning. SoT generates multiple low-cost intuitive thoughts, mimicking the parallel intuitions produced by System 1, and uses a confidence evaluator to cross-evaluate these thoughts. If conflicts arise, SoT invokes reflective reasoning of scaled-up LLMs, emulating System 2’s intervention, to rectify results. This model-agnostic framework can be implemented with various off-the-shelf LLMs. Experiments on six tasks show that SoT reduces API cost by 38.3%-75.1% while achieving state-of-the-art reasoning accuracy and solution diversity. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The authors want to help large language models work better without costing too much money. They created a new way called “Synergy of Thoughts” (SoT) that uses smaller language models to think of many ideas quickly, like humans do naturally. If these ideas don’t match, SoT asks the bigger language models to make it right. This method is flexible and can be used with different language models. It helps by reducing costs while still giving good answers. |
