Articles
Written as answers to questions people type when they want to make, vibe code, or choose a stack for a reinforcement learning environment, plus notes on where that work is heading.
Which AI is the best liar? (Dwarkesh Patel)
Dwarkesh Patel’s Blood on the Clock Tower benchmark: how frontier models scheme and bluff, and why he thinks LLMs are already building the RL environments used in their own training.
How can I make a reinforcement learning environment?
How to make a reinforcement learning environment: why RL environments matter for LLM progress, the observation-action-reward-done contract, and how Gymnasium, engines, and simulators compare.
How to vibe code a reinforcement learning environment
How to vibe code a reinforcement learning environment: why RL environments matter, how to name observation, action, reward, and done in a prompt, and how to check the draft by playing it.
Alternatives to Prime Intellect
Alternatives to Prime Intellect for RL environments: HUD, OpenEnv, Gymnasium, NeMo Gym, and OpenReward, compared by what you build, share, host, and train.
The economy will become an RL environment machine
Brendan Foody (Mercor CEO) on Lenny’s Podcast: why the economy may become an RL environment machine, who Mercor is, and what that means for building reinforcement learning environments.