Breaking
PLUrgent meeting of the president with the head of the Ministry of National Defense and the head of the General StaffCNAXA Wealth and Capital were involved in violating the banking law and attracted more than 600 million yuan. The inspection and dispatch troops were divided into 10 routes to search and interview 8 people.TRDrone attack from Russia on a minibus in Ukraine: 5 deadRUUkraine received a severe blow to the economy after losing control of the Dnieper-Bug estuaryINTLU.S. Imposes New Visa Restrictions on South African OfficialsDEVon der Leyen is promoting “associated” EU membership for CanadaDEVon der Leyen announces minimum age for social mediaDEEU wants to legally restrict social media access for minorsTRThe father who told ChatGPT about his plan to kill his son in Brazil was reported to the policeTRPrice manipulation and fake receipt operation: They paid the farmers low and made it seem highPLUrgent meeting of the president with the head of the Ministry of National Defense and the head of the General StaffCNAXA Wealth and Capital were involved in violating the banking law and attracted more than 600 million yuan. The inspection and dispatch troops were divided into 10 routes to search and interview 8 people.TRDrone attack from Russia on a minibus in Ukraine: 5 deadRUUkraine received a severe blow to the economy after losing control of the Dnieper-Bug estuaryINTLU.S. Imposes New Visa Restrictions on South African OfficialsDEVon der Leyen is promoting “associated” EU membership for CanadaDEVon der Leyen announces minimum age for social mediaDEEU wants to legally restrict social media access for minorsTRThe father who told ChatGPT about his plan to kill his son in Brazil was reported to the policeTRPrice manipulation and fake receipt operation: They paid the farmers low and made it seem high
BackOpenAI首席執行官山姆·阿爾特曼:世界應相信AI公司會做正確的事
OpenAI首席執行官山姆·阿爾特曼:世界應相信AI公司會做正確的事
Developing
BBC中文1 hour agoTech7 min readChinaView translation

OpenAI首席執行官山姆·阿爾特曼:世界應相信AI公司會做正確的事

在舊金山的一場會議上,OpenAI、Meta與英偉達的負責人紛紛表示支持行業自我監管,並強調安全與創新並重。

Quick Look

OpenAI首席執行官山姆·阿爾特曼在舊金山會議上表示,公眾應相信AI公司會做正確的事。此前有研究員警告AI可能毀滅人類,引發對該行業加強審查的呼聲,但多位科技巨頭高管仍主張進行自我監管而非政府介入。

AI-generated summary

Why It Matters

AI發展風險引發擔憂,前Anthropic研究員聲稱AI可能毀滅人類。科技領袖對自我監管與政府介入展開辯論。

Font size

OpenAI首席執行官山姆·阿爾特曼(Sam Altman;奧特曼)認為,儘管公眾對人工智能(artificial intelligence, AI;人工智慧)發展的各種風險的擔憂日益加劇,但人們應該對他的公司和其他同行公司更有信心,相信它們在 AI 開發上會做正確的選擇。

週二(9月15日)在舊金山(San Francisco)舉行的一場會議上,阿爾特曼表示:「世界應該相信我們會做正確的事,因為那就是正確的事,我們深切感受到它的重要性。」

然而,阿爾特曼指出,人們對 AI 的擔憂是有道理的,因為這些工具的能力發展迅速。

他說:「現在我們不用像以前那樣費腦筋,就能想像到這件事會如何出錯。我認為全世界對此感到恐懼是有道理的。」

阿爾特曼的言論是在他出席由軟件公司 Salesforce 主辦的年度會議時發表的。

這是自上週一則帖子在網絡上引發熱議以來,他首次公開發言。該帖子的作者是一名已辭職的人工智能公司 Anthropic 研究員。這名研究員聲稱,如果不加以控制,人工智能可能會在 2020 年代末毀滅所有人類。

其他一些人工智能公司的高層和專家回應稱,他們同意這個觀點,但並沒有解釋他們是如何得出結論的,也沒有解釋 AI 究竟是如何毀滅所有人類。

這些說法引發的熱議,近日促使外界對人工智能開發加強審視。作為回應,Anthropic 首席執行官達里奧·阿莫代伊(Dario Amodei)呼籲所有人工智能開發放緩步伐,並敦促各國政府對該行業加以監管。

這篇文章獲得了阿爾特曼的讚賞,谷歌(Google)旗下 DeepMind 聯合創始人德米斯·哈薩比斯(Demis Hassabis),以及社交媒體平台 X 和人工智能助手 Grok 的所有者埃隆·馬斯克(Elon Musk)也對此表示讚賞。

到週二為止,越來越多人工智能行業領袖表態支持行業自我監管,而非由政府介入。

阿爾特曼表示,他認為像自己公司這樣的人工智能企業,基本上有能力進行自我監管。

阿爾特曼說:「我們會把這件事做好,我對我們公司和整個行業有能力安全地做到這一點非常有信心。」他還表示,他確信這些公司會「讓對齊與安全遠遠領先於能力發展」,如果做不到,他們就會「放慢速度或停下來」。

在阿爾特曼發言之後,Meta 首席執行官扎克伯格(Mark Zuckerberg)在 X 上寫道,每一家人工智能實驗室都有能力、也有動機「採取自己的行動」,去開發以安全為設計原則的人工智能工具和模型。

扎克伯格寫道:「任何不重視對齊問題的實驗室都會落後。」「各實驗室若其模型造成損害,將面臨重大法律責任,因此它們同樣有強烈的動機去防止這種情況發生。」

英偉達(Nvidia;輝達)負責人黃仁勳(Jensen Huang)在週二同一場會議上也表示,是否發布新版本的相關技術,應由人工智能公司自行決定,而不應由外部力量來管理。

黄仁勳說:「我們不需要新的法律或法規。」他還表示,不應該在人工智能產品的創新速度和安全性之間做出「虛假的取捨」。

英偉達是全球市值最高的公司,其利潤因市場對其生產的人工智能運算晶片的旺盛需求而大幅增長。

黄仁勳說:「安全至關重要。不過,安全本質上是一個工程問題。」

他還表示,如果任何時候一家人工智能公司的負責人對自家產品缺乏信心,就應該選擇不發布該產品。

黄仁勳說:「這是顯而易見的做法⋯⋯盡可能全速前進,但只要你感覺到機構已經失控,就應該暫停下來。」

一些業內人士認為,近期人工智能恐慌情緒被誇大了,而且正被人借題發揮,為行業製造炒作熱度。

在會議期間,阿爾特曼向數百名商界人士發表講話,告訴他們應該使用人工智能工具來輔助工作,但同時也需要這些工具來保護他們的企業免受潛在的人工智能網路攻擊。

把人工智能行業完全交由其高管自我監管的想法,在業內一些人看來並不可取。

Anthropic 高管兼聯合創始人傑克·克拉克(Jack Clark)週一(14日)告訴BBC,放任人工智能成為一個「完全不受監管的行業」,是「拿著巨大風險在擲骰子」。

Logical Intelligence 公司首席運營官帕特里克·希爾曼(Patrick Hillman)週二指出,人們對科技公司真正為公眾利益做任何事都缺乏信心。這家公司的董事長是被視為人工智能行業泰斗的楊立昆(Yann LeCun)。

希爾曼說:「如今,美國人最不信任的機構可能只有矽谷了,其次才是華盛頓。我在矽谷和華盛頓都工作過,也都生活過我可以向你保證,這種不信任都是它們自己招來的。」

他還說:「如果你們自己也認為正在打造的東西是危險的,那就拿出實際行動,證明你們準備停止做哪些事情。」

OpenAI 和 Anthropic 的領導人表示,他們最近已開始努力達成某種行業範圍內的安全協議。

阿莫代伊在週二的會議上短暫露面時表示,Anthropic 目前正與業內其他公司就致力於提高 AI 工具和開發的安全標準和檢查進行「對話」。

What to Watch

AI outlook — possibilities, not facts

  • Anthropic與業內其他公司將繼續進行安全標準檢查的對話

    Likely · Within months

Open Questions

  • AI公司將如何具體落實安全標準?
  • 各國政府是否會強行推動行業監管?

Related Topics

This article was originally published by BBC中文.

Related Stories

New concerns about information security: Publicly available physical world data are collected and analyzed on a large scale by AI
Tech·

New concerns about information security: Publicly available physical world data are collected and analyzed on a large scale by AI

Research by TrendAI, a subsidiary of Trend Micro, pointed out that Chinese manufacturers use public physical world data such as flights, ships, and radios to conduct large-scale AI fusion analysis to restore sensitive military and corporate actions, forming an emerging information security risk called "open network entity telemetry."

自由时报
4 min read
Douyin announces the promotion of a "trustworthy and friendly" community building framework to systematically manage online rumors and online violence
Tech·

Douyin announces the promotion of a "trustworthy and friendly" community building framework to systematically manage online rumors and online violence

On September 15, Douyin announced at the 2026 National Cyber ​​Security Publicity Week Forum that it will promote the "Trusted Community Guardian Plan" and the "Friendly Community Co-construction Plan" to systematically manage issues such as rumors, AIGC abuse, and online violence through AI technology and rule construction.

中国新闻网
8 min read
Fii Industrial and Hongteng deepen cooperation, focusing on 800G, 1.6T and optoelectronic integration technology
Tech·

Fii Industrial and Hongteng deepen cooperation, focusing on 800G, 1.6T and optoelectronic integration technology

Zheng Hongmeng, Chairman of Industrial Fii, announced that Industrial Fii will deepen cooperation with Hongteng Precision, from upstream and downstream procurement to R&D and bottom-level co-creation, to jointly promote 800G, 1.6T and photoelectric integration technology to meet the global outbreak of AI computing power demand and interconnection bandwidth and energy consumption challenges.

自由时报
3 min read
Lu Songqing, Chairman of Hongteng Precision Technology: Data center efficiency in the AI era is more important than computing power
Developing·

Lu Songqing, Chairman of Hongteng Precision Technology: Data center efficiency in the AI era is more important than computing power

Lu Songqing, chairman of Hongteng Precision Technology, said that with the rapid development of AI, the focus of the industry has shifted from the pursuit of stronger computing power to maximizing the efficiency of data centers. He emphasized that every watt of power, data exchange and connection affects AI productivity, and proposed the concept of Large Scale, including Scale to Production and Scale with Ecosystem, to stably and reliably introduce innovative technologies to the global market.

自由时报
2 min read
More on this topic人工智能