Logan Kilpatrick(@OfficialLoganK)

What % of global AI token spend do you think is people running evals?

5.0内容质量
What % of global AI token spend do you think is people running evals?

TL;DR · AI 摘要

全球AI算力支出中用于模型评估的比例极低，用户猜测可能低于0.01%。

核心要点

  • 模型评估消耗的token占比可能低于0.01%
  • 开发者对评估重要性存在认知与实践的矛盾
  • 当前AI算力使用中评估环节被严重低估

结构提纲

按章节快速跳转。

  1. §问题提出

    Logan Kilpatrick质疑全球AI算力支出中评估占比

  2. ·用户反馈

    多数用户认为评估占比低于1%甚至0.01%

  3. ›行业现状

    实际算力使用中评估环节被严重低估

思维导图

用一张图看清主题之间的关系。

查看大纲文本（无障碍 / 无 JS 友好）
  • AI算力使用现状
    • 评估占比问题
      • 用户猜测<0.01%
      • 行业认知偏差
    • 算力分配
      • 训练主导
      • 评估被忽视

金句 / Highlights

值得收藏与分享的关键句。

#AI#算力#模型评估#token
打开原文

Logan Kilpatrick on X: "What % of global AI token spend do you think is people running evals?"

![Image 2: Google AI Studio](https://twitter.com/GoogleAIStudio) @OfficialLoganK What % of global AI token spend do you think is people running evals?

12:32 AM · Aug 20, 2026198.5K Views

<1% I think we use more tokens saying “ thank you”

![Image 5: OpenRouter](https://twitter.com/OpenRouter) @jjacky Aug 20 fairly confident it's less than 0.01% makes me sad, b/c evals are so important

![Image 7: Palantir](https://twitter.com/PalantirTech) @chadwahl Aug 20 Precisely as many as people eating their broccoli routinely