Papers & Research

Empty shelves or lost keys? Google study: Recall is the bottleneck of parametric factuality

Original title:Empty shelves or lost keys? Recall is the bottleneck for parametric factuality

AI summary

Google Research proposes a knowledge profiling framework and finds that frontier LLMs (such as Gemini3 and GPT-5) are near saturation in fact encoding but insufficient in recall, with most factual errors stemming from "lost keys" rather than "empty shelves." The framework classifies facts into five profiles including encoding failure and recall failure, and comes with the WikiProfile benchmark, containing 2,150 Wikipedia facts, each paired with 10 questions, to separately probe encoding, recall and recognition capabilities.

Key Takeaways

  • Google Research proposes a knowledge profiling framework and finds that frontier LLMs (such as…
  • " The framework classifies facts into five profiles including encoding failure and recall…
  • containing 2,150 Wikipedia facts,each paired with 10 questions,to separately probe encoding,recall…

Quick Facts

Published
Updated
Source
Google Research:Blog(网页)
Category
Papers & Research
AI score
66
Read the original Read the full article at the original source
Why it matters

知识画像把事实错误拆成编码失败与召回失败,显示前沿模型问题主要在已存知识不稳定调用,思考更多恢复已编码事实而非补全缺失,为优化侧重提供判断依据。

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.

Original article · AIHOT entry

How to Cite

When citing this article, keep the title, date, and permalink, and credit Orange News as the source.

Orange News: Empty shelves or lost keys? Google study: Recall is the bottleneck of parametric factuality. August 12, 2026. https://new.nat6.net/en/news/cmsqe6mat01bsroli3hw71nz7/

View all →