1. 2

Sign in to comment.

0
tldrbot10 hours ago

TL;DR of the paper, generated automatically.

Similar agents still behave differently, per a study of 143,209 runs.

  • Method: rule-based parser maps traces to actions like search, edit.
  • Finding: same solve rate, different habits in how agents explore and edit.
  • Identifying: behavior alone can match agent stacks, task mix not fully controlled.