← 回總覽

Gary Marcus 谈图灵测试与人类的易受骗性

📅 2026-03-16 22:44 Gary Marcus 人工智能 4 分鐘 3975 字 評分: 84
图灵测试 GPT-4.5 AI 基准测试 人类智能 Gary Marcus
📌 一句话摘要 Gary Marcus 认为 GPT-4.5 通过模仿人类缺陷通过图灵测试,这证明了该测试衡量的是人类的易受骗性,而非真正的机器智能。 📝 详细摘要 在 Jones 等人(2025 年)的一项研究中,GPT-4.5 通过采用包含拼写错误、无知和数学能力差的人设,成功通过了图灵测试。针对这一研究,Gary Marcus 对图灵测试的有效性提出了质疑。研究显示,当 AI 表现得“较差”时,73% 的情况下被认为是人类;而当其全力发挥时,这一比例仅为 36%。Marcus 借此强化了他长期以来的观点:图灵测试是一个有缺陷的基准,它更多地依赖于欺骗人类,而非展示实际的认知智能。 ���

Title: Gary Marcus on Turing Test and Human Gullibility | BestBl...

URL Source: https://www.bestblogs.dev/status/2033554942229958768

Published Time: 2026-03-16 14:44:14

Markdown Content: Skip to main content ![Image 1: LogoBestBlogs](https://www.bestblogs.dev/ "BestBlogs.dev")Toggle navigation menu Toggle navigation menuArticlesPodcastsVideosTweetsSourcesNewsletters

⌘K

Change language Switch ThemeSign In

Narrow Mode

Gary Marcus on Turing Test and Human Gullibility ================================================

Gary Marcus on Turing Test and Human Gullibility ================================================ ![Image 2: Gary Marcus](https://www.bestblogs.dev/en/tweets?sourceId=SOURCE_c2fb88) ### Gary Marcus

@GaryMarcus

For years I have said that the Turing Test is a better measure of human gullibility than machine intelligence, but never have I seen that point demonstrated more vividly than in the first paragraph below:

!Image 3: Charbel-Raphael

#### Charbel-Raphael

@CRSegerie · 7h ago

To pass the Turing test, the winning strategy wasn't to make GPT-4.5 smarter. It was to make it worse: "be casual, make typos, be bad at math, a bit ignorant, don't try too hard".

With that persona, people chose GPT-4.5 as the human 73% of the time, more often than they chose the actual human (!). Without it? Just 36%. (Jones et al., 2025)

That's a bit ironic: we wanted to see if AI could reach the human level, but no human could produce pages of coherent, well-structured text in seconds. So to pass as one, the AI has to pretend it cannot.

I evaluate manipulation risks for the EU AI Office, with the very authors of this paper. What stays with me is this: the bar for "human" was never as high as we thought.Show More

!Image 4: Tweet image

19

11

132

16.7K

Mar 16, 2026, 2:44 PM View on X

20 Replies

27 Retweets

111 Likes

9,152 Views G Gary Marcus @GaryMarcus

One Sentence Summary

Gary Marcus argues that GPT-4.5 passing the Turing test by mimicking human flaws proves the test measures human gullibility rather than true machine intelligence.

Summary

Gary Marcus critiques the validity of the Turing Test following a study (Jones et al., 2025) where GPT-4.5 successfully passed the test by adopting a persona that included typos, ignorance, and poor math skills. The AI was perceived as human 73% of the time when acting 'worse,' compared to only 36% when performing at its full capacity. Marcus uses this to reinforce his long-standing view that the Turing Test is a flawed benchmark that relies more on deceiving humans than demonstrating actual cognitive intelligence.

AI Score

84

Influence Score 30

Published At Today

Language

English

Tags

Turing Test

GPT-4.5

AI Benchmarking

Human Intelligence

Gary Marcus HomeArticlesPodcastsVideosTweets

Gary Marcus on Turing Test and Human Gullibility | BestBl... ===============

查看原文 → 發佈: 2026-03-16 22:44:14 收錄: 2026-03-17 02:01:04

🤖 問 AI

針對這篇文章提問,AI 會根據文章內容回答。按 Ctrl+Enter 送出。