按年被提到
20232023 · 48 次2026
大模型与语言模型›对话式大模型·arXiv:2303.12712 · cs.CL

Sparks of Artificial General Intelligence: Early experiments with GPT-4

通用人工智能的火花:GPT-4 的早期实验
Sébastien Bubeck, Varun Chandrasekaran, Ronen Eldan, Johannes Gehrke, Eric Horvitz, Ece Kamar, Peter Lee, Yin Tat Lee, Yuanzhi Li, Scott Lundberg, Harsha Nori, Hamid Palangi 等 14 人 · 2023

微软研究院对早期 GPT-4 的大规模能力测试与 AGI 讨论

开始阅读 ↗PDFalphaXiv 讨论
被推荐 · 52 人首次 2023-03-23 · 最近 2026-08-03
52不同的人
14活跃月份
53讨论串
3,206被引(OpenAlex)

作者考察了 OpenAI 开发中的早期 GPT-4,认为它与 ChatGPT、PaLM 等同属表现出比以往模型更通用智能的新一批 LLM。他们展示 GPT-4 无需特别提示就能处理数学、编程、视觉、医学、法律、心理学等领域的新颖难题,表现常接近人类水平,并认为它可被视为不完整的早期 AGI;同时着重讨论其局限,以及超越逐词预测的新范式的可能需要。

HN 为什么提它

HN 上的人在争论 LLM 是否只是压缩或统计模仿时反复拿它作证据,引用其中的数学、代码、TikZ 独角兽和 SVG 作图等例子;有人说它改变了自己对 AI 进展的看法,也有人只认为它令人印象深刻而非 AGI。

实证研究难度 2 / 5约 300 分钟有时效性
读之前最好知道
1大语言模型(LLM)基本概念
2提示(prompting)的基本用法
3常见编程与数学题的基础
4对 AGI 讨论的基本了解

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
强烈推荐visarga · 2023-03

“after reading the Sparks of Artificial General Intelligence paper I am convinced it is proto-AGI.”

强烈推荐:读完后确信 GPT-4 是 proto-AGI
ChatGPT Plugins原评论 ↗
强烈推荐WonderBuilder · 2023-03

“(an incredible, incredible document) and check out Section 10.1”

称这是一份极其出色的文档,推荐看 10.1 节
ChatGPT Plugins原评论 ↗
强烈推荐ux-app · 2023-03

“my whole world view changed when I read Sparks of Artificial General Intelligence”

称读完这篇论文,自己的整个世界观都变了
BlenderGPT: Use commands in English to control Blender with OpenAI's GPT-4原评论 ↗
强烈推荐Springtime · 2024-07

“certainly shifted my views last year”

强烈推荐:去年读它确实改变了自己对 LLM 进展的看法
Large language models have developed a higher-order theory of mind原评论 ↗
推荐meowkit · 2023-03

“Kinda mind blowing among some of the many experiments they published in here:”

推荐,称其中不少实验令人惊叹,例如 GPT-4 生成 SVG 图像
BlenderGPT: Use commands in English to control Blender with OpenAI's GPT-4原评论 ↗
推荐kzrdude · 2023-03

“I guess that's a good place to start to answer your question. I don't think this is AGI, but the paper is full of examples where GPT-4 works well and does impressive stuff.”

推荐:不认为这是 AGI,但论文里有大量 GPT-4 表现出色的例子
看全部 20 条 →
推荐Version467 · 2023-04

“I can recommend the Sparks of AGI paper[2] and the toolformer paper[3].”

推荐 Sparks of AGI 论文和 Toolformer 论文
Saying Goodbye to GitHub – Ersei 'N Stuff原评论 ↗
推荐lhl · 2023-04

“If you (or anyone else is) interested in the topic, I'd highly recommend giving some of these a look:”

强烈建议想了解这个话题的人看看这类资料,包括这篇 GPT-4 能力的论文
The LLama Effect: Leak Sparked a Series of Open Source Alternatives to ChatGPT原评论 ↗
推荐pmoriarty · 2023-04

“I strongly disagree with that view, and encourage reading the "Sparks of Artificial General Intelligence" paper[1] and watching the associated video[2]”

强烈反对“只是压缩”的看法,建议读这篇论文并看相关视频
33TB of text data for a 1T-parameter model原评论 ↗
推荐jiggawatts · 2023-04

“Early experiments with GPT-4" better. Link: Sparks of Artificial General Intelligence”

贴出链接,更喜欢它“Early experiments with GPT-4”这个副标题
Emergent Abilities of Large Language Models原评论 ↗
推荐codermind · 2023-04

“you will know it’s a baby AGI.”

称读完就会知道它是个婴儿 AGI
AGI Is Not Chess Bots – They Will Replace Your Jobs原评论 ↗
推荐seanhunter · 2023-05

“This paper is startling in terms of a world model and how it seems to emerge in a specific example in the step from GPT3.5 to 4.”

称论文令人震惊:从 GPT-3.5 到 4 的某个例子里似乎涌现出了世界模型
Just Calm Down About GPT-4 Already原评论 ↗
推荐nopinsight · 2023-05

“The Sparks of AGI paper provides much more evidence and examples: Sparks of Artificial General Intelligence”

推荐:它提供了多得多的 GPT-4 能力证据和例子
Statement on AI Risk原评论 ↗
推荐vczf · 2023-07

“Check out the tikz unicorn drawing example from the paper”

推荐看论文里 GPT-4 用 TikZ 画独角兽的例子
Experiencing Decreased Performance with ChatGPT-4原评论 ↗
推荐w1nk · 2023-08

“Read this paper from a team at Microsoft Research:”

推荐微软研究院这篇论文,约 150 页例子展示 GPT-4 对世界的理解
AI Isn’t Good Enough原评论 ↗
批评pabo · 2023-04

“they give the impression that GPT-4 is an even bigger step for AI than I think is realistically true.”

Sparks of AGI? – Analyzing GPT-4 and the Latest GPT/LLM Models [sentdex]原评论 ↗
批评nradov · 2023-05

“What the paper actually shows is that even intelligent people have a bias towards perceiving patterns in randomness”

Geoffrey Hinton leaves Google and can now speak freely about his AI concern原评论 ↗
批评nybsjytm · 2024-07

“has nothing whatsoever to do with Socrates or the way he argued”

批评论文里的“苏格拉底式对话”与苏格拉底及其论辩方式毫无关系,只是普通聊天
I am starting an AI+Education company原评论 ↗
批评alkyon · 2025-08

“What else is he known for in the field of computer science?”

Claim: GPT-5-pro can prove new interesting mathematics原评论 ↗
批评romaniv · 2026-05

“100% marketing, 0% science.”

The last six months in LLMs in five minutes原评论 ↗