The short version
An open-source experiment where an RL agent writes RL training jobs.
From the HackerLinks archive
ai-trains-ai
An open-source experiment where an RL agent writes RL training jobs.
At a glance:
First seen:2026-07-14
Last seen:2026-07-14
Times seen:1
Website:github.com
Why it caught our attention
A concrete, reproducible nested-RL prototype with a clear cost and limitation discussion.
Where it surfaced on Hacker News
2026-07-14
Editorial paraphrase
Its author described 1,750 training jobs, hidden-eval feedback, and limits to small agentic tasks on the Prime-RL stack.
Original threadShow HN: I RL-trained an agent that trains models with RL (for ~$1.3k)