{"slug": "the-surprising-effectiveness-of-deleting-weights-in-llm-post-training", "title": "The Surprising Effectiveness of Deleting Weights in LLM Post-Training", "summary": "Zeroing fewer than 0.05% of a pretrained large language model's weights, with no other modification, matches full fine-tuning at scale across three LLM post-training regimes, including reinforcement learning with verifiable rewards, according to the researchers behind the finding. The result suggests new capabilities can be acquired by changing only a tiny fraction of a model's parameters rather than retraining the whole network.", "body_md": "How little of a pretrained large language model (LLM) has to change for it to acquire new capabilities? We find that zeroing fewer than 0.05% of its weights, with no other modification, matches full fine-tuning (FFT) at scale across three LLM post-training regimes: reinforcement learning with verifi", "url": "https://wpnews.pro/news/the-surprising-effectiveness-of-deleting-weights-in-llm-post-training", "canonical_source": "https://aiflash.com/news/149752/", "published_at": "2026-11-11 13:00:05+00:00", "updated_at": "2026-11-11 13:23:40.954300+00:00", "lang": "en", "topics": ["large-language-models", "machine-learning", "ai-research", "artificial-intelligence"], "entities": [], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/the-surprising-effectiveness-of-deleting-weights-in-llm-post-training", "markdown": "https://wpnews.pro/news/the-surprising-effectiveness-of-deleting-weights-in-llm-post-training.md", "text": "https://wpnews.pro/news/the-surprising-effectiveness-of-deleting-weights-in-llm-post-training.txt", "jsonld": "https://wpnews.pro/news/the-surprising-effectiveness-of-deleting-weights-in-llm-post-training.jsonld"}}