Researchers are developing new methods to improve the long-horizon decision-making capabilities of language agents. One approach, RELACE, uses retrospective likelihood scoring to assess actions based on their plausibility in both original and outcome-augmented contexts, demonstrating significant improvements on tasks like ALFWorld and WebShop. Another framework, Stateless Language Agents (SLAs), separates research state from agent conversations, allowing stateless agents to focus on specific tasks assigned by an advisor, leading to better performance with fewer tokens. Additionally, a study on action calibration suggests that explicitly linking past actions to their outcomes and using learned calibrators can enhance agent performance by preventing ineffective repetition and improving decision-making. AI
IMPACT These advancements in language agent decision-making could lead to more capable and efficient AI systems for complex, long-term tasks.
RANK_REASON Multiple research papers published on arXiv detailing new methodologies for improving language agent decision-making over long horizons.
- Action Calibration
- arXiv
- Computation and Language
- Hugging Face
- language agents
- assurance ledger
- $au^2$-bench
- language-model agents
- Organising Trajectory Evidence for Language-Model Agent Assurance: Fragments, Methods, and the Residual
- rule checkers
- ALFWorld
- GRPO
- HCAPO
- Qwen2.5-1.5B-Instruct
- Qwen2.5-7B-Instruct
- Stateless Language Agents
- WebShop
AI-generated summary · Google Gemini · from 5 sources. How we write summaries →