Open Models Almanac
Methodspost-trainingagentic post-training

single mention · post-training

Autonomous Execution Tasks

Autonomous Execution Tasks (AET) is a reinforcement-learning environment paradigm for training long-horizon agents via verify-in-the-loop optimization.1 Agents are given a goal, constraints, and verification interfaces but no reference trajectories, and must autonomously plan, use tools, and recover from errors, with rewards based on an independent verifier's evaluation of the final state.2

How this definition was written

Summarised by glm-5.3-flash from 1 source; every sentence is checked against the quote it cites.

  1. Autonomous Execution Tasks (AET), an environment paradigm that trains long-horizon agent intelligence through verify-in-the-loop optimization — k3 tech report
  2. Rewards are grounded in the verifier’s evaluation of the final environment state rather than the agent’s self-reported completion — k3 tech report

Also called Autonomous Execution Tasks (AET).

Filed under post-training › agentic post-training