---
title: "Discussion on \"Can we really scale RL?\" | Hashnode"
description: "Discussion on \"Can we really scale RL?\". Yes and No. LLM reasoning research is just a big pile of math. We stir the math every once in a while, and it starts doing crazy stuff. For months the community has argued that RL post-training just polishes ideas an LLM already had. ProRL politely s..."
url: "https://hashnode.com/posts/can-we-really-scale-rl/68456aebee7fd0185efda965"
---

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

Discussion

# [Can we really scale RL?](https://blog.ritwikraha.dev/can-we-really-scale-rl?utm_source=hashnode&utm_medium=feed)

Yes and No. LLM reasoning research is just a big pile of math. We stir the math every once in a while, and it starts doing crazy stuff. For months the community has argued that RL post-training just polishes ideas an LLM already had. ProRL politely s...

Jun 8, 2025 · 11 min read · blog.ritwikraha.dev  
Updated Jun 11, 2025  
Read the article

-   [#reinforcement-learning](https://hashnode.com/tag/reinforcement-learning)
-   [#machine-learning](https://hashnode.com/tag/machine-learning)
-   [#llm](https://hashnode.com/tag/llm)
-   [#large-language-models](https://hashnode.com/tag/large-language-models)
-   [#claudeai](https://hashnode.com/tag/claudeai)
-   [#gpt](https://hashnode.com/tag/gpt)
-   [#rlhf](https://hashnode.com/tag/rlhf)

## Written by

[Ritwik Raha](https://hashnode.com/@ritwikraha)

@ritwikraha

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

## Responses

Markdown supported · ⌘↵ to post

No responses yet.
