人工智慧將在 2025 年重塑世界，但大多數人並沒有意識到

如果 2024 年是人工智慧融入主流的一年，那麼 2025 年將是人工智慧開始徹底重塑它的一年。

2024 was the year AI went mainstream. But just wait until 2025.

2024 年是人工智慧成為主流的一年。但要等到 2025 年。

Most people are still blissfully unaware of just how different the world is about to look. But for those immersed in the tech space, the pace of change is exhilarating—and overwhelming.

大多數人仍然幸福地沒有意識到世界即將變得多麼不同。但對於那些沉浸在科技領域的人來說，改變的步伐令人振奮，而且勢不可擋。

OpenAI’s latest models, o3 and o3-mini, exemplify this transformative moment. These models aren’t just smarter chatbots; they’re problem solvers, excelling at reasoning tasks like complex coding, advanced mathematics, and even scientific challenges. O3 represents a new breed of AI that not only processes data but thinks—a shift that pushes us closer to artificial general intelligence (AGI).

OpenAI 的最新型號 o3 和 o3-mini 體現了這一變革時刻。這些模型不僅僅是更聰明的聊天機器人；他們是問題解決者，擅長複雜編碼、高等數學甚至科學挑戰等推理任務。 O3 代表了一種新型人工智慧，它不僅能處理數據，還能思考──這項轉變讓我們更接近通用人工智慧 (AGI)。

The o3 models achieve breakthroughs in reasoning through test-time compute, a mechanism allowing the AI to ‘ponder’ over complex problems before delivering a response. This innovative leap sets the stage for AI to tackle questions that demand deeper understanding, moving beyond rote predictions to reasoning capabilities that challenge the boundaries of what’s possible with machines.

o3 模型透過測試時計算在推理方面取得了突破，這種機制允許人工智慧在做出回應之前「思考」複雜的問題。這項創新飛躍為人工智慧解決需要更深入理解的問題奠定了基礎，超越死記硬背的預測，發展到挑戰機器可能性界限的推理能力。

o3 is better than 99.95% of programmers Source: X

o3 優於 99.95% 的程式設計師 資料來源：X

From Transforming Industries to Challenging Geopolitical Norms

從轉型產業到挑戰地緣政治規範

But these advances aren’t happening in a vacuum. They’re part of a broader narrative of AI-driven disruption, from transforming industries to challenging geopolitical norms.

但這些進步並不是憑空發生的。它們是人工智慧驅動的顛覆的更廣泛敘述的一部分，從產業轉型到挑戰地緣政治規範。

As AI becomes increasingly integral to our daily lives, its infrastructure demands have skyrocketed. AI is no longer just a software story. It’s about infrastructure—massive, tangible infrastructure. To make AI work, we need chips, electricity, and interconnects at an unprecedented scale.

隨著人工智慧越來越融入我們的日常生活，其基礎設施需求也隨之激增。人工智慧不再只是一個軟體故事。它與基礎設施有關——大規模、有形的基礎設施。為了讓人工智慧發揮作用，我們需要前所未有的晶片、電力和互連。

Consider this: OpenAI’s ambitions include building data centers that demand as much power as entire cities. We’re not just talking about a few racks of servers; we’re talking about infrastructure projects on par with major metropolitan utilities.

想想看：OpenAI 的雄心壯志包括建立需要與整個城市一樣多電力的資料中心。我們談論的不僅僅是幾個伺服器機架；我們談論的是與主要大都市公用事業同等的基礎設施項目。

This is a CAPEX boom that doesn’t just benefit Silicon Valley engineers but electricians, construction workers, and local businesses. AI is making atoms great again, driving wealth creation across a broader swath of society than any previous tech wave.

這是資本支出的繁榮，不僅有利於矽谷工程師，還有電工、建築工人和當地企業。人工智慧正在讓原子再次變得偉大，推動社會財富的創造比之前的任何科技浪潮都要廣泛。

While social media enriched a small subset of programmers and influencers, AI’s hunger for physical infrastructure is creating opportunities for the concrete mixer, the electrician, and the local HVAC shop. This isn’t just a digital revolution; it’s a physical one, with economic ripples that extend far beyond traditional tech hubs.

雖然社群媒體豐富了一小部分程式設計師和有影響力的人，但人工智慧對實體基礎設施的渴望正在為混凝土攪拌機、電工和當地的暖通空調商店創造機會。這不僅是一場數位革命，更是一場數位革命。這是一種物理性的影響，其經濟影響遠遠超出了傳統科技中心的範圍。

Test-Time Compute

測試時計算

But this isn’t just a story of growth. It’s a story of constraints—and cracks are already forming. Scaling laws, once gospel in the AI world, are facing practical limits.

但這不僅僅是一個成長的故事。這是一個充滿限制的故事，而且裂痕已經形成。規模法則曾經是人工智慧世界的福音，但現在正面臨實際限制。

Pretraining these massive models requires staggering amounts of data and power, pushing current infrastructure and resources to their breaking point. The hyperscalers like Google, Amazon, and Microsoft are scrambling to secure energy sources, from renewables to nuclear, to keep their data centers running.

預先訓練這些龐大的模型需要大量的數據和電力，將目前的基礎設施和資源推向極限。谷歌、亞馬遜和微軟等超大規模企業正爭先恐後地確保從再生能源到核能等能源的安全，以保持資料中心的運作。

OpenAI’s o3 introduces a promising workaround: test-time compute. This allows models to ‘ponder’ more complex questions, using more resources for tougher problems and optimizing performance. It’s an elegant solution, but it’s only the beginning.

OpenAI 的 o3 引進了一個很有前景的解決方法：測試時計算。這使得模型能夠「思考」更複雜的問題，使用更多的資源來解決更棘手的問題並優化效能。這是一個優雅的解決方案，但這只是一個開始。

The implications of these advancements are staggering. AI’s reach will extend into every industry, from medicine to agriculture, manufacturing to science. It’s poised to create orders of magnitude more wealth than previous technologies by impacting everything, everywhere, all at once.

這些進步的影響是驚人的。人工智慧的影響力將擴展到各個行業，從醫學到農業，從製造到科學。透過同時影響一切、無所不在，它有望創造比以前的技術多幾個數量級的財富。

Critics might call this another tech bubble, but AI has the two critical ingredients to prove them wrong: massive capital investment and unparalleled generality. It’s not just a new tool; it’s a General Purpose Technology (GPT) on the scale of electricity, reshaping the fundamental fabric of human productivity.

批評者可能會稱之為另一個科技泡沫，但人工智慧有兩個關鍵因素可以證明他們是錯的：大量的資本投資和無與倫比的通用性。它不僅僅是一個新工具；它是一個新工具。它是一種電力規模的通用技術（GPT），重塑了人類生產力的基本結構。

“This chart is pretty insane, it’s the current AGI progression in passing the ARC test. o3 already exceeded the average human decision-making scores, where will OpenAI’s o4 be in 6 months?” Source: X

「這張圖表相當瘋狂，它是當前 AGI 通過 ARC 測試的進度。 o3已經超過了人類決策的平均得分，6個月後OpenAI的o4會在哪裡？來源：X

Still, key questions loom large as we enter 2025. Are we hitting walls in pretraining? How do we solve reasoning at scale? When will the nation-state race to AGI begin?

儘管如此，隨著我們進入 2025 年，一些關鍵問題變得越來越突出。我們如何解決大規模推理問題？民族國家的 AGI 競賽何時開始？

OpenAI’s breakthroughs, like o1’s chain-of-thought reasoning and scaling inference compute, offer glimpses of answers. But the road ahead will test the limits of innovation, infrastructure, and human coordination.

OpenAI 的突破，例如 o1 的思想鏈推理和擴展推理計算，提供了一些答案。但前方的道路將考驗創新、基礎設施和人類協調的極限。

The U.S. is leading the AI race, with regulatory frameworks emerging and partnerships between AI labs and defense sectors strengthening. Meanwhile, other nations are lagging, though that’s unlikely to last.

美國在人工智慧競賽中處於領先地位，監管框架不斷湧現，人工智慧實驗室與國防部門之間的合作關係也得到加強。與此同時，其他國家正在落後，儘管這種情況不太可能持續下去。

Europe’s regulatory-driven approach and China’s puzzling hesitation in prioritizing AI could shift as the race heats up. The geopolitical implications are profound.

隨著競爭的加劇，歐洲的監管驅動方式和中國在優先考慮人工智慧方面令人困惑的猶豫可能會改變。其地緣政治影響是深遠的。

National strategies around AI are emerging as critical differentiators, with some countries already positioning themselves as AI superpowers. In the U.S., regulatory efforts are balancing innovation with control, while strategic partnerships between AI labs and defense sectors signal an acknowledgment of AI’

圍繞人工智慧的國家戰略正在成為關鍵的差異化因素，一些國家已經將自己定​​位為人工智慧超級大國。在美國，監管工作正在平衡創新與控制，而人工智慧實驗室和國防部門之間的戰略合作夥伴關係則標誌著對人工智慧的認可。