Papers

12094 papers

TALM: Tool Augmented Language Models

Summary pending...

ART: Automatic multi-step reasoning and tool-use for large language models

Summary pending...

TMLR2025

Diff-Instruct++: Training One-step Text-to-image Generator Model to Align with Human Preferences

This paper presents Diff-Instruct++, a one-step text-to-image generator model designed to better align generated images with human preferences. By addressing the limitations of existing models, it enhances the quality and relevance of generated images based on textual descriptions.

text-to-image generationhuman preferencesmachine learning
ICLR2025

Efficient Reinforcement Learning with Large Language Model Priors

This paper presents a method for enhancing reinforcement learning (RL) by incorporating large language model (LLM) priors, which helps improve the efficiency and effectiveness of learning agents. The approach leverages the rich contextual understanding of LLMs to guide decision-making in RL tasks, making it a significant advancement in the field.

reinforcement learninglarge language modelssample efficiency
ICLR2025

Rewarding Progress: Scaling Automated Process Verifiers for LLM Reasoning

This paper presents a method for scaling automated process verifiers to enhance the reasoning capabilities of large language models (LLMs). By rewarding incremental progress in verification tasks, the approach aims to improve LLM performance in complex reasoning scenarios, which is crucial for their deployment in real-world applications.

automated verificationlarge language modelsreinforcement learning
ICLR2025

Efficient Online Reinforcement Learning Fine-Tuning Need Not Retain Offline Data

This paper presents a method for fine-tuning reinforcement learning agents online without the need to retain offline data. This approach enhances efficiency and reduces the storage burden associated with traditional methods.

reinforcement learningonline learningdata efficiency
ICLR2025

Direct Post-Training Preference Alignment for Multi-Agent Motion Generation Model Using Implicit Feedback from Pre-training Demonstrations

This paper presents a method for aligning multi-agent motion generation models with human preferences using implicit feedback derived from pre-training demonstrations. This approach enhances the model's ability to generate more desirable behaviors in robotic applications, making it significant for improving autonomy in complex environments.

multi-agent systemsmotion generationrobotics
EMNLP2023

ChatCoT: Tool-Augmented Chain-of-Thought Reasoning on Chat-based Large Language Models

The paper introduces ChatCoT, a method that enhances chat-based large language models by integrating tool-augmented reasoning through chain-of-thought techniques. This approach improves the models' ability to perform complex tasks by leveraging external tools for better decision-making and problem-solving.

Tool UseLarge Language ModelsReasoning Techniques
EMNLP2024

LLM-A\*: Large Language Model Enhanced Incremental Heuristic Search on Path Planning

This paper presents LLM-A*, an approach that integrates large language models into incremental heuristic search for path planning. It aims to enhance the efficiency and effectiveness of finding optimal paths in complex environments, which is crucial for applications in robotics and AI navigation.

path planningheuristic searchlarge language models

Understanding the planning of LLM agents: A survey

This survey paper explores the planning mechanisms employed by large language model (LLM) agents, highlighting their capabilities and limitations. Understanding these planning strategies is crucial for improving LLM performance in complex tasks and applications.

LLMplanningsurvey
NeurIPS2023

ToolkenGPT: Augmenting Frozen Language Models with Massive Tools via Tool Embeddings

This paper introduces ToolkenGPT, a method that enhances frozen language models by integrating them with a variety of external tools through the use of tool embeddings. This advancement allows for improved task performance by leveraging specialized functionalities of these tools, making language models more versatile.

Tool UseLanguage ModelsNeurIPS 2023
2025

A Review of Prominent Paradigms for LLM-Based Agents: Tool Use (Including RAG), Planning, and Feedback Learning

This paper explores how to improve the coherence of understanding across different repositories. It highlights the importance of consistent data interpretation for better information retrieval and knowledge management.

data coherenceinformation retrievalknowledge management
NeurIPS2023

HuggingGPT: Solving AI Tasks with ChatGPT and its Friends in Hugging Face

This paper introduces HuggingGPT, a framework that integrates ChatGPT with various models from Hugging Face to tackle a range of AI tasks. It highlights the potential of combining different AI models to enhance task performance and flexibility.

AI IntegrationChatGPTHugging Face

A Survey on Large Language Model based Autonomous Agents

This paper surveys the development and application of large language model-based autonomous agents, highlighting their capabilities and challenges. It is important as it provides a comprehensive overview of how these agents are transforming various fields and identifies future research directions.

large language modelsautonomous agentssurvey