Sign In
insight - LLM-based Agents
대형 언어 모델 기반 에이전트를 위한 긍정적 경험 반영: Sweet&Sour 접근 방식
텍스트 기반 게임 환경에서 대형 언어 모델(LLM) 기반 에이전트의 성능을 향상시키기 위해 긍정적 경험과 부정적 경험 모두를 활용하는 새로운 반영 기법인 Sweet&Sour를 제안한다.
インタラクティブなテキスト環境におけるエージェントのためのポジティブエクスペリエンスの考察：成功と失敗からの学習
LLMベースのエージェントにおける自己反省の際に、従来の失敗からの学習だけでなく、成功体験も活用することで、より効果的な学習とパフォーマンスの向上が見込める。
CAAP: A Modular LLM-Based Agent for Solving Computer Tasks Using Only Visual Input and Keyboard/Mouse Actions
This paper introduces CAAP, a novel LLM-based agent that can solve computer tasks by mimicking human behavior, relying solely on visual input from the screen and executing actions through keyboard and mouse commands.
EPO: Using Environment Feedback to Train Hierarchical Large Language Model Agents for Long-Horizon Decision-Making
Hierarchical LLM agents trained with a novel Environment Preference Optimization (EPO) method, which leverages multimodal environment feedback to generate reward and preference signals, achieve state-of-the-art performance on long-horizon decision-making tasks in the ALFRED benchmark.
Products
© 2024 by Linnk AI