Sign In

PieArena: Frontier Language Agents Achieve MBA-Level Negotiation Performance and Reveal Novel Behavioral Differences

Created by
  • Haebom
Category
Empty

μ €μž

Chris Zhu (Department of Statistics,Data Science, Yale University), Sasha Cui (Department of Statistics,Data Science, Yale University), Will Sanok Dufallo (Department of Philosophy, Yale University), Runzhi Jin (School of Law, University of California, Berkeley), Zhen Xu (Bloomberg), Linjun Zhang (Department of Statistics, Rutgers University), Daylian Cain (Yale School of Management)

πŸ’‘ κ°œμš”

λ³Έ μ—°κ΅¬λŠ” LLM의 ν˜‘μƒ λŠ₯λ ₯을 ν‰κ°€ν•˜κΈ° μœ„ν•΄ MBA ν˜‘μƒ μˆ˜μ—…μ„ 기반으둜 ν•œ λŒ€κ·œλͺ¨ 닀쀑 μ—μ΄μ „νŠΈ ν˜‘μƒ 벀치마크인 PieArenaλ₯Ό κ°œλ°œν–ˆμŠ΅λ‹ˆλ‹€. μ΅œμ²¨λ‹¨ μ–Έμ–΄ λͺ¨λΈ(GPT-5)은 MBA 학생듀과 λ™λ“±ν•˜κ±°λ‚˜ κ·Έ μ΄μƒμ˜ ν˜‘μƒ μ„±λŠ₯을 보여 AGI μˆ˜μ€€μ˜ λŠ₯λ ₯을 μž…μ¦ν–ˆμŠ΅λ‹ˆλ‹€. λ˜ν•œ, 쀑간 및 ν•˜μœ„ ν‹°μ–΄ λͺ¨λΈμ˜ μ„±λŠ₯ ν–₯상을 κ°€μ Έμ˜¨ joint-intentionality μ—μ΄μ „νŠΈ μŠ€μΊν΄λ”©μ˜ 효과λ₯Ό λΆ„μ„ν•˜κ³ , κΈ°μ‘΄ λ²€μΉ˜λ§ˆν¬μ—μ„œ κ°„κ³Όλ˜μ—ˆλ˜ 기만, 계산 μ •ν™•μ„±, μ§€μ‹œ μ€€μˆ˜, μΈμ‹λœ ν‰νŒ λ“± 닀차원적인 행동 ν”„λ‘œν•„μ„ κ³΅κ°œν–ˆμŠ΅λ‹ˆλ‹€.

πŸ”‘ μ‹œμ‚¬μ  및 ν•œκ³„

β€’
μ΅œμ²¨λ‹¨ μ–Έμ–΄ λͺ¨λΈμ€ 이미 κ³ μœ„ν—˜ 경제 ν™˜κ²½μ— 적용될 수 μžˆμ„ 만큼 지적 및 심리적 λŠ₯λ ₯을 κ°–μΆ”κ³  μžˆμŠ΅λ‹ˆλ‹€.
β€’
Joint-intentionality μŠ€μΊν΄λ”©μ€ ν•˜μœ„ ν‹°μ–΄ λͺ¨λΈμ˜ ν˜‘μƒ λŠ₯λ ₯ ν–₯상에 νš¨κ³Όμ μž…λ‹ˆλ‹€.
β€’
λͺ¨λΈμ˜ 강건성(robustness)κ³Ό μ‹ λ’°μ„±(trustworthiness) 뢀쑱은 μ—¬μ „νžˆ ν•΄κ²°ν•΄μ•Ό ν•  과제둜 λ‚¨μ•„μžˆμŠ΅λ‹ˆλ‹€.
πŸ‘