TechPulseTelegram
โ† PrevMon, Sep 7Next โ†’

๐Ÿค– AI Agent10 items

Latest AI agent research papers, analysis, and improvement insights

semantic_scholarresearch

Spawn Freely, Act Sparingly: Progressive Risk Vesting for Recursive LLM-Agent Trees

๋‹ค์ค‘ ์—์ด์ „ํŠธ์˜ ํ–‰๋™ ๊ถŒํ•œ์„ ๋‹จ๊ณ„์ ์œผ๋กœ ๋ถ€์—ฌํ•˜๋Š” ์œ„ํ—˜ ๊ด€๋ฆฌ ํ”„๋ ˆ์ž„์›Œํฌ

semantic_scholarresearch

Understanding Stage-Wise Utility-Risk Trade-offs in LLM Agent Memory

Agent ๋ฉ”๋ชจ๋ฆฌ ์„ค๊ณ„ ๋‹จ๊ณ„์—์„œ ์œ ์šฉ์„ฑ๊ณผ ๋ณด์•ˆ์˜ ๊ท ํ˜•์„ ๋ถ„์„ํ•˜๋Š” ํ‰๊ฐ€ ๋„๊ตฌ

semantic_scholarresearch

EDGE: Error Dependency Graph-Guided Multi-Error Attribution in Multi-Agent LLM Systems

๋‹ค์ค‘ ์—์ด์ „ํŠธ ์‹œ์Šคํ…œ์˜ ๋ณตํ•ฉ ์—๋Ÿฌ ๊ฐ„ ์ข…์†์„ฑ์„ ๋ถ„์„ํ•˜๋Š” ๋””๋ฒ„๊น… ํ”„๋ ˆ์ž„์›Œํฌ

semantic_scholarresearch

Bilevel Coordinated Reflection: A Game-Theoretic Approach to Multi-Agent LLM Systems

์˜ค์ผ€์ŠคํŠธ๋ ˆ์ดํ„ฐ-์›Œ์ปค ์ƒํ˜ธ์ž‘์šฉ์„ ๊ฒŒ์ž„ ์ด๋ก ์œผ๋กœ ๋ชจ๋ธ๋งํ•œ ๋‹ค์ค‘ ์—์ด์ „ํŠธ ์กฐ์œจ ํ”„๋ ˆ์ž„์›Œํฌ

semantic_scholarresearch

SoK: When Safe Agents Fail Together: The Security of Multi Agent LLM Systems

๋‹ค์ค‘ ์—์ด์ „ํŠธ ์‹œ์Šคํ…œ์˜ ๊ฒฝ๊ณ„ ๊ฐ„ ์ •๋ณด ํ๋ฆ„์œผ๋กœ ์ธํ•œ ๋ณด์•ˆ ์œ„ํ—˜ ์ฒด๊ณ„ํ™”

semantic_scholarresearch

MASkills: Continual Skills Optimization for Multi-Agent LLM Systems

๊ฒฝํ—˜์œผ๋กœ๋ถ€ํ„ฐ ๊ตฌ์กฐํ™”๋œ ์ ˆ์ฐจ์  ์ง€์‹์„ ์ง€์†์ ์œผ๋กœ ํ•™์Šตํ•˜๋Š” ๋‹ค์ค‘ ์—์ด์ „ํŠธ ํ”„๋ ˆ์ž„์›Œํฌ

semantic_scholarresearch

When Guardrails Look Effective: Construct Validity Failures in LLM Agent Commerce Evaluation

LLM ์—์ด์ „ํŠธ ์‹œ๋ฎฌ๋ ˆ์ด์…˜์˜ ํ‰๊ฐ€ ์ฒ™๋„๊ฐ€ ์‹ค์ œ ํ–‰๋™์„ ๋ฐ˜์˜ํ•˜์ง€ ๋ชปํ•˜๋Š” ์œ„ํ—˜์„ ๋ถ„์„

semantic_scholarresearch

GlossoGen: Emergent Language in Complex Multi-Agent LLM Interactions

๋ถ€๋ถ„ ์ •๋ณด ์ƒํƒœ์—์„œ ๋‹ค์ค‘ ์—์ด์ „ํŠธ์˜ ์–ธ์–ด ์ง„ํ™”์™€ ์•ˆ์ „์„ฑ ๋ชจ๋‹ˆํ„ฐ๋ง ์—ฐ๊ตฌ

semantic_scholarresearch

NS-Copilot: An LLM-Driven Agent System for Autonomous Neuroscience Analysis

์‹ ๊ฒฝ๊ณผํ•™ ์—ฐ๊ตฌ๋ฅผ ์œ„ํ•ด ์ด์งˆ์ ์ธ ์‹ ๊ฒฝ ๋ชจ๋ธ์„ ํ†ตํ•ฉํ•˜๋Š” LLM ์—์ด์ „ํŠธ ์‹œ์Šคํ…œ

blog_openairesearch

Research acceleration: The view inside OpenAI

์ฝ”๋”ฉ ์—์ด์ „ํŠธ์˜ ์—ฐ๊ตฌ ์ƒ์‚ฐ์„ฑ, ์‹คํ—˜ ์†๋„, ๋ณต์žก๋„ ๋“ฑ ์‹ค์ „ ํ™œ์šฉ ๋ฐ์ดํ„ฐ ๊ณต๊ฐœ