Original Paper: https://arxiv.org/abs/2308.10379

By: Bilgehan Sel, Ahmad Al-Tawaha, Vanshaj Khattar, Ruoxi Jia, Ming Jin

Abstract:

Current literature, aiming to surpass the "Chain-of-Thought" approach, often resorts to an external modus operandi involving halting, modifying, and then resuming the generation process to boost Large Language Models' (LLMs) reasoning capacities. This mode escalates the number of query requests, leading to increased costs, memory, and computational overheads. Addressing this, we propose the Algorithm of Thoughts -- a novel strategy that propels LLMs through algorithmic reasoning pathways, pioneering a new mode of in-context learning. By employing algorithmic examples, we exploit the innate recurrence dynamics of LLMs, expanding their idea exploration with merely one or a few queries. Our technique outperforms earlier single-query methods and stands on par with a recent multi-query strategy that employs an extensive tree search algorithm. Intriguingly, our results suggest that instructing an LLM using an algorithm can lead to performance surpassing that of the algorithm itself, hinting at LLM's inherent ability to weave its intuition into optimized searches. We probe into the underpinnings of our method's efficacy and its nuances in application.

Summary Notes

Enhancing AI Problem-Solving with the Algorithm of Thoughts (AoT)

In the fast-moving world of Artificial Intelligence (AI), Large Language Models (LLMs) are getting closer to thinking and solving problems like humans.

Yet, making them work efficiently on complex tasks is still a big challenge. Recent approaches like Chain of Thought (CoT) and Self-Consistency have made some progress but require a lot of computing power.

The Algorithm of Thoughts (AoT) is a new strategy aimed at making LLMs explore ideas more efficiently, promising to improve performance without needing as much computational effort.

Current Methods in AI Reasoning

First, let's review the existing techniques:

• Standard Prompting: This basic approach uses simple input-output prompts, which can result in less accurate answers.
• Chain of Thought (CoT): CoT breaks down reasoning step-by-step but is limited to linear thinking.
• Self-Consistency: This method looks for a consensus among multiple reasoning paths, increasing accuracy but also computational needs.
• Tree of Thoughts (ToT): ToT examines different reasoning paths using an external tree search, which also requires significant resources.

Each method has pushed LLMs forward but also shown the need for more efficient problem-solving capabilities.

Algorithm of Thoughts (AoT): A Step Forward

AoT introduces a new way by integrating algorithmic reasoning directly into the LLM's process of generating responses. It simulates human problem-solving techniques, such as recursive thinking and backtracking, in a more computationally friendly manner. AoT's key features include:

1. Breaking Down Problems: It simplifies complex problems into smaller, more manageable tasks.
2. Generating Solutions: Utilizes the LLM's ability to generate creative solutions for these smaller tasks.
3. Evaluating and Backtracking: Continuously assesses and refines solutions, improving the overall process.

AoT Performance Evaluation

AoT was tested on complex tasks like the game of 24 and mini-crosswords, where it outperformed CoT, CoT-SC, and ToT by providing accurate answers with fewer queries. This not only shows AoT's potential to improve LLM problem-solving but also its efficiency in using computational resources.

The Impact of AoT on Future Problem-Solving

Incorporating AoT into LLMs represents a significant leap in AI's ability to handle complex reasoning tasks more like humans. By lowering computational requirements and improving problem-solving capabilities, AoT opens new possibilities for AI applications in various fields, including healthcare and data analysis.

Conclusion

The Algorithm of Thoughts (AoT) is a promising development in making LLMs more efficient and capable problem solvers.

By embedding algorithmic reasoning into the generative process of LLMs, AoT not only cuts down on computational needs but also significantly enhances their ability to tackle complex tasks. As AI continues to advance, methods like AoT will be key to overcoming current limitations, leading to more sophisticated AI systems.

For AI engineers in enterprise settings, AoT offers a groundbreaking approach to creating innovative, efficient, and cost-effective AI solutions, unlocking the full potential of LLMs in problem-solving.