{"slug": "paper-summary-the-matthew-effect-in-rl", "title": "Paper Summary: The Matthew Effect in RL", "summary": "A paper titled \"Learning to Solve Hard Problems in RL for LLMs by Never Giving Up\" proposes countering the Matthew Effect in reinforcement learning, where LLMs improve far more on easy problems they already solve than on harder problems with lower solve rates, by allocating training time dynamically based on problem difficulty. The approach targets the post-training stage, where models are commonly tuned with reinforcement learning methods such as GRPO after supervised fine-tuning on a large corpus.", "body_md": "[Learning to Solve Hard Problems in RL for LLMs by Never Giving Up](https://arxiv.org/pdf/2609.13443v1) discusses an approach for countering what the authors call the *Matthew Effect*, the tendency for the LLM to improve much more on easy problems that it is already good than harder problems that have a lower solve rate. Their approach is to allocate training time dynamically based on the difficulty of the problem.\n\nAfter an LLM has been pretrained on a large corpus of supervised fine-tuning (SFT) data, it is common to post-train [using reinforcement learning methods like GRPO](./posts/notes-on-deepseek-r1/). This allows the model to improve on tasks with verifiable rewards and even exceed the performance of the original SFT data.", "url": "https://wpnews.pro/news/paper-summary-the-matthew-effect-in-rl", "canonical_source": "https://blog.lukesalamone.com/posts/matthew-effect/", "published_at": "2026-09-26 09:36:55+00:00", "updated_at": "2026-09-28 05:48:52.423469+00:00", "lang": "en", "topics": ["machine-learning", "large-language-models", "ai-research", "artificial-intelligence"], "entities": ["GRPO", "DeepSeek-R1"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/paper-summary-the-matthew-effect-in-rl", "markdown": "https://wpnews.pro/news/paper-summary-the-matthew-effect-in-rl.md", "text": "https://wpnews.pro/news/paper-summary-the-matthew-effect-in-rl.txt", "jsonld": "https://wpnews.pro/news/paper-summary-the-matthew-effect-in-rl.jsonld"}}