Glossary · Term

PyTerm

← all terms

Definition

Plain language

A test set of Python scripting tasks for AI agents working at a command line.

As stated in the literature

A Python-scripting benchmark for terminal agents where command outputs are tightly coupled to the agent's own actions; used in ECHO's verifier-free experiments, where reward-free environment-prediction training improved task success.

Why it matters: Because each command's result depends directly on the agent's own actions, it serves as a clean testbed for whether agents can learn from predicting what their commands will do.

For example, a task might ask an agent to write and run a short Python script at the command line and check that its output is correct.

Heard on the show

“On a Python scripting benchmark called PyTerm, pass rate goes up by ten percentage points.”
Episode 084 — Terminal Agents Get Free Supervision From The Tokens We've Been Throwing Away

Mentioned in 1 episode

  1. 084
    Terminal Agents Get Free Supervision From The Tokens We've Been Throwing Away

Related terms