Portland, OR, USA - May 2, 2025: Assorted AI apps, including ChatGPT, Gemini, Claude, Perplexity, Meta AI, Microsoft Copilot, and Grok, are seen on the screen of an iPhone.
Cover 面對技術瓶頸與龐大資本支出壓力,科技界近日開始重新評估前沿 AI 模型的開發節奏與商業轉化效率。(圖片:Getty Images)
Portland, OR, USA - May 2, 2025: Assorted AI apps, including ChatGPT, Gemini, Claude, Perplexity, Meta AI, Microsoft Copilot, and Grok, are seen on the screen of an iPhone.

Elon Musk 倡議美中 AI 巨頭互相測試模型,Anthropic 等領袖亦呼籲放緩開發。這究竟是出於安全考量,還是面對技術瓶頸與資本泡沫的市場策略?

人工智能(AI)領域的科技競賽近年一直以驚人速度推進,然而近期多位業界重量級人物卻罕有地發出「減速」的訊號。特斯拉(Tesla)兼 xAI 創辦人 Elon Musk 在洛杉磯舉行的 All-In 峰會上提出一項前瞻性建議,主張包括 xAI、OpenAI、Google、Meta 以及數家頂尖中國人工智能企業在內的全球巨頭,應建立統一的測試平台,在產品正式推出前讓競爭對手互相進行安全性交叉測試(Cross-testing)。與此同時,Anthropic 執行長 Dario Amodei 近期亦發表文章倡議合理控制前沿技術的推進節奏,而 OpenAI 執行長 Sam Altman 等人亦對相關安全風險表達關注。這股由行業領袖發起的「主動放緩」思潮,表面上旨在建立跨國界的安全防護網,避免自我審查的盲點;然而,在資本市場對 AI 龐大資本支出回報率產生疑慮、以及技術迭代邊際效益放緩的背景下,這究竟是出於對人類文明安全的負責任抉擇,抑或是面對研發瓶頸與商業壓力下的策略性調整?

延伸閱讀:Gen.T 2026 先鋒領袖名單揭曉!他們是塑造香港未來的先驅

Tatler Asia
DAVOS, SWITZERLAND - JANUARY 22: Business person Elon Musk delivers a speech during the World Economic Forum Annual Meeting in Davos, Switzerland, on January 22, 2026. (Photo by Harun Ozalp/Anadolu via Getty Images)
Above Elon Musk 在洛杉磯 All-In 峰會上倡議,美中兩國 AI 巨頭應建立統一平台,互相測試模型的安全性。(圖片:Getty Images)
DAVOS, SWITZERLAND - JANUARY 22: Business person Elon Musk delivers a speech during the World Economic Forum Annual Meeting in Davos, Switzerland, on January 22, 2026. (Photo by Harun Ozalp/Anadolu via Getty Images)

越界交叉測試的合規挑戰

Elon Musk 提出的「交叉測試」概念,核心價值在於透過競爭對手的視角找出模型內部測試難以察覺的安全漏洞。然而在實際執行層面上,該倡議面臨著極高的商業機密與地緣政治屏障。首先,前沿 AI 模型的權重(Model Weights)與架構屬於科技巨頭的核心資產,開放競爭對手存取無疑增加了技術抄襲與逆向工程的風險。其次,將美中兩國的 AI 企業納入同一測試體系,在現行晶片出口管制與數據跨境傳輸的監管框架下存在相當難度。雖然中美兩國政府與部分研究機構均對 AI 安全風險保持關注,但缺乏具約束力的國際監督機制,使這項互相審查的機制目前仍停留在構想階段。

Tatler Asia
SAN FRANCISCO, CALIFORNIA - SEPTEMBER - 15: Anthropic CEO Dario Amodei, left, speaks with Salesforce CEO Marc Benioff during the keynote address at Salesforce's Dreamforce conference at the Moscone Center on September 15, 2026 in San Francisco, California. Dreamforce is an annual event that highlights the company's technologies and encourages professional networking. (Photo by Benjamin Fanjoy/Getty Images)
Above Anthropic 執行長 Dario Amodei 發表文章倡議合理控制 AI 推進節奏,強調前沿安全的重要性。(圖片:Getty Images)
SAN FRANCISCO, CALIFORNIA - SEPTEMBER - 15: Anthropic CEO Dario Amodei, left, speaks with Salesforce CEO Marc Benioff during the keynote address at Salesforce's Dreamforce conference at the Moscone Center on September 15, 2026 in San Francisco, California. Dreamforce is an annual event that highlights the company's technologies and encourages professional networking. (Photo by Benjamin Fanjoy/Getty Images)

研發成本與技術瓶頸

開發前沿模型所面臨的客觀技術與財務限制,亦是產業步調出現變化的重要原因。隨著模型規模持續擴大,訓練下一代超大規模模型所需的算力成本、電力消耗以及高質素數據資源均呈指數級上升。業界數據顯示,單次頂級模型的訓練成本已高達數億美元,而模型在推理能力與泛化能力上的提升幅度,相比早期階段已顯現出邊際效益遞減的趨勢。高昂的資本支出與尚未完全成熟的商業化盈利模式,使不少科技企業在追求純粹模型規模之餘,轉而將資源投放於提升架構效率、優化特定應用場景及確保安全性上。這種實質上的開發節奏調整,客觀上為企業提供了優化資源配置的緩衝期。

延伸閱讀:7 位亞裔創辦人領軍,打造美國估值破十億美元的科技初創巨頭

Tatler Asia
LONDON, ENGLAND - SEPTEMBER 16: A placard reading 'Pause AI the risk is too high' is seen on the ground on Whitehall near Downing Street on September 16, 2026 in London, England. PAUSE AI are asking the government to listen to the people and consider the dangers of AI. They say there are real-life harms of AI, including amplifying biases, spreading widespread misinformation, consolidating power among elite tech companies, and straining public infrastructure, housing, and utilities. Recent incidents they hig
Above 隨著歐盟《人工智慧法案》等合規條款陸續生效,全球監管政策大幅提高了科技企業的研發與營運成本。(圖片:Getty Images)
LONDON, ENGLAND - SEPTEMBER 16: A placard reading 'Pause AI the risk is too high' is seen on the ground on Whitehall near Downing Street on September 16, 2026 in London, England. PAUSE AI are asking the government to listen to the people and consider the dangers of AI. They say there are real-life harms of AI, including amplifying biases, spreading widespread misinformation, consolidating power among elite tech companies, and straining public infrastructure, housing, and utilities. Recent incidents they hig

地緣政治與監管政策

產業倡議放緩的背景,亦包含全球監管環境與地緣政治的複雜影響。歐盟《人工智慧法案》(AI Act)的陸續實施,以及各國對前沿科技審查的強化,使得合規成本顯著增加。儘管美國政府官員曾表示政府應鼓勵私營部門創新,並擔心過度限制可能影響競爭優勢,但科技巨頭主動提出安全測試標準,客觀上有助於在政府推出更嚴格的強制性法規前,確立產業自律的框架。此外,將中國企業納入安全討論,反映出科技領袖意識到 AI 安全屬於全球性議題。總括而言,AI 領域今後的發展,必須在創新速度、商業轉化與監管合規三者之間取得平衡,同時亦是現階段科技企業決策層無法迴避的課題。

Topics