Study Infers Personality Traits of 80 LLM Agents From 345K Trajectories

A new paper introduces the A-B-D method, which infers behavioral traits of 80 LLM agents bottom-up from 345,000 real trajectories across 50 task types, without relying on self-reports or LLM judges. Findings show clear trait differences, with Claude ranking most planful and GPT-5.5 the least active.

2026-10-23 ~ 2026-10-23 · 2 related posts