---
title: "Discussion on \"RL in the Pre-train Space: Why Training on P(y) Beats Training on P(y|x)\" | Hashnode"
description: "Discussion on \"RL in the Pre-train Space: Why Training on P(y) Beats Training on P(y|x)\". A new paper shows that reinforcement learning directly on the marginal distribution unlocks reasoning capabilities that standard RLVR can never reach."
url: "https://hashnode.com/posts/rl-in-pretrain-space-training-py-beats-pyx/69e101636ef723c0870126bb"
---

[Feed](https://hashnode.com/feed) /

Discussion

# [RL in the Pre-train Space: Why Training on P(y) Beats Training on P(y|x)](https://theagentstack.hashnode.dev/rl-in-pretrain-space-training-py-beats-pyx?utm_source=hashnode&utm_medium=feed)

A new paper shows that reinforcement learning directly on the marginal distribution unlocks reasoning capabilities that standard RLVR can never reach.

Apr 16 · 7 min read · theagentstack.hashnode.dev  
Read the article

-   [#artificial-intelligence](https://hashnode.com/tag/artificial-intelligence)
-   [#deep-learning](https://hashnode.com/tag/deep-learning)
-   [#machine-learning](https://hashnode.com/tag/machine-learning)

## Written by

[Manas Vardhan](https://hashnode.com/@manasvardhan)

@manasvardhan · Los Angeles, USA

  
[View profile](https://hashnode.com/@manasvardhan)

I write philosophy, literature and tech

## Responses

Markdown supported · ⌘↵ to post

No responses yet.
