Thought Buffer (BoT): A New Thought-Augmented Reasoning AI Approach to Improve Accuracy, Efficiency, and Robustness of LLMs

Remarkable performance on different reasoning tasks has been demonstrated by several large language models (LLMs), such as GPT-4, PaLM, and LLaMA. To further increase the functionality and performance of LLMs, there are more effective stimulation methods and an increase in model size, both of which improve reasoning performance. The approaches are classified as follows: (i) methods that rely on a single query to complete the reasoning process, such as those used for rapid engineering; (ii) methods that use multiple LLM queries to produce different plausible reasoning paths, breaking complex problems into smaller ones; Examples of this type of reasoning include Least to Great, ToT, and GoT.

However, there are limitations to both types of methods:

• It is impractical to manually design single-query reasoning systems on a task-by-task basis because they typically rely on prior assumptions or relevant examples of reasoning processes.
• Multi-query reasoning systems are computationally intensive because they recursively expand reasoning paths to find a unique intrinsic structure for each task.
• Both single-query and multi-query reasoning systems are limited by their reasoning structures and examples. They fail to derive general, high-level patterns or thoughts from previously completed tasks, which could be useful in improving efficiency and accuracy when solving similar problems.

Introducing a novel approach to address these limitations, a team of researchers from Peking University, UC Berkeley, and Stanford University have developed the Buffer of Thoughts (BoT). This innovative and flexible framework for thought-augmented reasoning is designed to improve the accuracy, efficiency, and resilience of LLM reasoning across a wide range of tasks. A key component of BoT is the metabuffer, a small library that stores a set of high-level generalizable ideas (thinking templates) drawn from various problem-solving procedures. These thinking templates can be reused for other tasks, facilitating effective thought-powered reasoning and configured with a specific reasoning structure.

BoT is designed to be stable and scalable, so the team included a buffer manager to update the metabuffer dynamically. In this way, the capacity of the metabuffer effectively increases as more work is done. The three main benefits of this approach are:

1. Improved precision: By using shared thinking templates, high-level thoughts can be instantiated to adaptively address various tasks. This eliminates the need to build reasoning structures from the beginning, dramatically improving reasoning accuracy.
2. Streamlined Reasoning: By directly utilizing informational historical reasoning structures, proposed thought-powered reasoning could streamline reasoning processes and eliminate cumbersome multiple query procedures.
3. BoT's approach to retrieving and instantiating thoughts mirrors human brain processes, enhancing LLMs' ability to consistently solve similar problems. This improves the robustness of the model, and when applied to various tasks, experimental results show that BoT significantly improves accuracy, efficiency, and resilience. These practical benefits make BoT a promising tool to improve the performance of LLMs in real-world applications.

Researchers build a buffer manager to extract ideas from different solutions and improve the capacity of the metabuffer as more tasks are completed. They conduct extensive experiments on ten difficult tasks that require a lot of reasoning. With an average cost of only 12% of multi-query request approaches, BoT outperforms previous SOTA methods by 51% on Checkmate in One, 11% on Game of 24, and 20% on Geometric Shapes.

The proposed approach greatly improves the accuracy while keeping the reasoning efficient and robust. However, when it comes to problems that require human ingenuity, the method has little to offer, because these problems often do not have a precise thinking template. Additionally, the resulting thought templates might not be of the best quality if BoT uses a less robust model to initialize the metabuffer. This is because the weaker model has restricted reasoning and instruction-following capabilities. Taken together, the following are the paths forward that BoT reveals: 1. Create an open domain system, such as an agent model, combining BoT with external resources. 2. optimize the distillation of thinking templates, which could greatly improve their ability as templates for increasingly complicated activities.

Review the Paper and GitHub. All credit for this research goes to the researchers of this project. Also, don't forget to follow us on twitter.com/Marktechpost”>twitter. Join our Telegram channel, Discord channeland LinkedIn Grabove.

If you like our work, you will love our Newsletter..

Don't forget to join our 44k+ ML SubReddit