按年被提到
20232023 · 8 次2026
AI 安全、伦理与社会影响›通用智能与意识·arXiv:2311.02462 · cs.AI

Levels of AGI for Operationalizing Progress on the Path to AGI

AGI 的分级:为通往 AGI 之路的进展建立可操作的度量
Meredith Ringel Morris, Jascha Sohl-Dickstein, Noah Fiedel, Tris Warkentin, Allan Dafoe, Aleksandra Faust, Clement Farabet, Shane Legg · 2023

按性能深度与通用广度给 AGI 分级的框架

开始阅读 ↗PDFalphaXiv 讨论
被推荐 · 13 人首次 2023-11-08 · 最近 2026-09-24
13不同的人
9活跃月份
17讨论串
138被引(OpenAlex)

论文提出一个对 AGI 模型及其前身的能力和行为进行分类的框架,引入关于性能、通用性和自主性的分级,为比较模型、评估风险和衡量进展提供共同语言。作者先分析现有 AGI 定义并提炼出一个有用的 AGI 本体应满足的六条原则,据此按能力的深度(性能)与广度(通用性)提出「AGI 分级」,讨论现有系统如何归入其中,以及未来基准测试面临的要求,最后讨论这些分级与自主性、风险等部署因素的关系。

HN 为什么提它

HN 上的人在争论「什么算 AGI」时反复引用它,认为这个由 DeepMind 提出的分级分类清晰、能让说法更具体,有人主张直接采用它,也有人拿它来给 ChatGPT 归类。

观点难度 2 / 5约 60 分钟经得起时间
读之前最好知道
1大语言模型的基本能力
2AI 基准测试(benchmark)的概念
3对 AGI 各类定义的大致了解
4AI 安全与风险的基本概念

HN 书评

推荐在前,同类按时间 · 每篇最多保留 20 条,统计用全部提及
推荐lucubratory · 2023-11

“Paper from two weeks ago by Google DeepMind putting together a very good definition”

推荐这篇 DeepMind 的论文,给出了很好的 AGI 定义
Sutskever: OpenAI board doing its mission to build AGI that benefits all原评论 ↗
推荐lucubratory · 2023-12

“there's a good DeepMind paper on a relatively concrete scaling system on the path to AGI:”

推荐这篇 DeepMind 论文,它给出了通往 AGI 路上相对具体的分级体系
Why are you so sure I'll lose?原评论 ↗
推荐theptip · 2024-03

“I think the DeepMind taxonomy is good; recognize that AGI is too broad a term, so add levels to make it specific.”

赞同 DeepMind 的分类:AGI 一词太宽泛,加上分级才具体
Elon Musk sues Sam Altman, Greg Brockman, and OpenAI [pdf]原评论 ↗
推荐theptip · 2024-06

“Here is a good one:”

推荐这篇好论文,它用严格明确的 AGI 分级取代宽泛的定义
Getting 50% (SoTA) on Arc-AGI with GPT-4o原评论 ↗
推荐theptip · 2025-05

“Can we just use Morris et al and move on with our lives?”

建议直接采用 Morris 等人的 AGI 分级定义,不要再争论
John Carmack talk at Upper Bound 2025原评论 ↗
引用ignoramous · 2023-11

“He and DeepMind claim AlphaGo is already AGI in a very narrow domain.”

引用它说明作者和 DeepMind 认为 AlphaGo 已是狭窄领域的 AGI
The deal to bring Sam Altman back to OpenAI has fallen apart原评论 ↗
看全部 15 条 →
引用kevindamm · 2023-12

“this was explicitly left aside in a recent paper that attempts to give measurement criteria for AGI[0]”

指出这篇给出 AGI 衡量标准的论文明确把具身性排除在外
Audiobox: Meta's new foundation research model for audio generation原评论 ↗
引用pmayrgundter · 2024-05

“DeepMind's Position paper "Levels of AGI for Operationalizing Progress on the Path to AGI" has a schema like this, where AGI capabilities are defined across 2 axes of Performance level X Generality (narrow vs general)”

引用它的框架:AGI 能力按表现水平与通用性(狭窄或通用)两个维度划分
Transformers Can Do Arithmetic with the Right Embeddings原评论 ↗
引用voidspark · 2025-05

“There can be levels of AGI. Google DeepMind have proposed a framework that would classify ChatGPT as "Emerging AGI".”

拿它说明 AGI 可以分级,Google DeepMind 的框架把 ChatGPT 归为“初现的 AGI”
Evolving OpenAI's Structure原评论 ↗
引用voidspark · 2025-05

“There can be levels of AGI. Google DeepMind have proposed a framework that would classify ChatGPT as "Emerging AGI".”

拿它说明 AGI 可以分级,Google DeepMind 的框架把 ChatGPT 归为“初现的 AGI”
Evolving OpenAI's Structure原评论 ↗
引用johnb231 · 2025-05

“Researchers at Google have proposed a classification scheme with multiple levels of AGI.”

说明 Google 研究者提出了把 AGI 分为多个级别的分类方案
John Carmack talk at Upper Bound 2025原评论 ↗
引用esafak · 2025-05

“See also: Levels of AGI for Operationalizing Progress on the Path to AGI”

作为参考补充:AGI 分级框架
How Does Claude 4 Think? – Sholto Douglas and Trenton Bricken原评论 ↗
引用simonw · 2025-06

“DeepMind: Levels of AGI for Operationalizing Progress on the Path to AGI”

引用 DeepMind 的 AGI 定义,强调元认知和学习新技能
Seven replies to the viral Apple reasoning paper and why they fall short原评论 ↗
引用theptip · 2025-10

“probably somewhere between expert and virtuoso AGI Levels of AGI for Operationalizing Progress on the Path to AGI”

借其分级把当前水平大致放在专家与大师级 AGI 之间
Andrej Karpathy – It will take a decade to work through the issues with agents原评论 ↗
引用theptip · 2026-09

“If you believe that capabilities will taper off exactly at human levels (i.e. "Competent AGI" from [1])”

借其“胜任级 AGI”分级,讨论能力是否恰好在人类水平后趋缓
Dynamic Abliteration: Non-Destructive Refusal Suppression via Engram Steering原评论 ↗
被提交为 story · 4 次