Thinking is recalling: how reasoning unlocks parametric knowledge in LLMs
Original title:Thinking to recall: How reasoning unlocks parametric knowledge in LLMs
AI summary
Google Research found that reasoning (chain-of-thought) helps large language models (LLMs) recall simple facts, even when these facts require no complex derivation. On Gemini-2.5 Flash and Pro and Qwen3-32B, enabling reasoning allows models to answer simple questions they otherwise could not answer directly, and pass@k shows the correct facts exist in the output distribution. The phenomenon is driven by two mechanisms: generated reasoning tokens act as a computational buffer, allowing the model to perform hidden computation to extract parametric knowledge; and related facts generated during reasoning serve as factual priming, helping the model activate the correct answer.
Key Takeaways
- Google Research found that reasoning (chain-of-thought) helps large language models (LLMs) recall…
- On Gemini-2.
- 5 Flash and Pro and Qwen3-32B,enabling reasoning allows models to answer simple questions they…
- The phenomenon is driven by two mechanisms: generated reasoning tokens act as a computational…
- and related facts generated during reasoning serve as factual priming,helping the model activate…
Quick Facts
- Published
- Updated
- Source
- Google Research:Blog(网页)
- Category
- Papers & Research
- AI score
- 69
发现一个反直觉现象,让模型推理能提升简单事实回忆,不是靠分步解题而是计算缓冲和事实预热,对理解模型知识召回机制很有启发,但中间步骤幻觉也带来风险。
Related entities
Source & copyright
This article is based on curated titles and AI summaries from the AIHOT public API, with attribution and links back. Original copyright remains with the original publishers and authors; this site does not mirror full texts.
How to Cite
When citing this article, keep the title, date, and permalink, and credit Orange News as the source.
Orange News: Thinking is recalling: how reasoning unlocks parametric knowledge in LLMs. June 24, 2026. https://new.nat6.net/en/news/cmqsbr8l701ndslfubzp0hnmi/