AoT

aot

Object

AoT (Atom of Thoughts) is an innovative reasoning framework that transforms complex reasoning processes into a Markov-style sequence of atomic questions. This enables more efficient and effective problem-solving in large language models (LLMs) during test-time scaling.

Features

  • Two-phase transition mechanism: decomposition and contraction of questions
  • Markov process approach that eliminates the need to maintain historical dependencies
  • Ability to represent complex reasoning as a series of independent, self-contained subquestions
  • Plug-in enhancement capability for existing test-time scaling methods
  • Resource optimization by focusing computational power on the current question state
  • Support for diverse reasoning scenarios including math, multi-choice, and multi-hop QA
  • Dependency-based directed acyclic graph representation of problem decomposition
  • Seamless integration with LangChain and other frameworks

Outcome

AoT significantly enhances LLM performance across multiple benchmarks while reducing computational waste. When applied to gpt-4o-mini on HotpotQA, it achieves an 80.6% F1 score, surpassing o3-mini by 3.4% and DeepSeek-R1 by 10.6%. By transforming reasoning into atomic states with memoryless transitions, AoT enables more efficient resource allocation and improved reasoning capabilities, making it particularly effective for complex problem-solving tasks that require multi-step reasoning.

Learn more about AoT →

View Next

AFlow

AFlow is an automated framework for generating and optimizing agentic workflows tailored to large language models (LLMs). This innovation eliminates the need for manual workflow design while enhancing performance and reducing associated costs.

Data Interpreter

Data Interpreter is an advanced LLM-based agent designed to solve complex data science problems end-to-end. It addresses the challenges of handling interconnected tasks, dynamic data adjustments, and domain-specific expertise requirements that traditional approaches struggle with. The system employs sophisticated hierarchical modeling to break down complex problems into manageable components while maintaining adaptability to real-time changes.