AI培訓的能源危機：與格雷格·奧蘇里（Greg Osuri）的權力下放能否提供解決方案？

人工智能培訓要求巨大的能量。格雷格·奧蘇里（Greg Osuri）提出權力下放，以解決緩解能源危機的解決方案。

The AI boom is here, but it's got a dirty little secret: it guzzles energy like a monster truck at a gas station. As AI models get bigger and brainier, the energy needed to train them is skyrocketing, raising concerns about an impending energy crisis. Greg Osuri, founder of Akash Network, is ringing the alarm bell, suggesting that the industry is underestimating the compute demands and environmental costs that come with it.

AI繁榮在這裡，但是它有一個骯髒的小秘密：它像加油站一樣笨拙的能量。隨著AI模型變得更大和腦，訓練它們所需的能量正在飆升，這引起了人們對即將發生的能源危機的擔憂。 Akash Network的創始人Greg Osuri正在敲響警鐘，這表明該行業低估了隨之而來的計算需求和環境成本。

The AI Energy Wall

AI能壁

Osuri points out that data centers, the powerhouses behind AI, already consume hundreds of megawatts of fossil fuel power. This trend could lead to soaring household power bills and millions of tons of new emissions annually. The concentration of fossil fuel use around data hubs even poses health risks. According to a Bloomberg report, wholesale electricity costs have surged 267% in five years in areas near data centers.

Osuri指出，數據中心是AI背後的強力，已經消耗了數百兆瓦的化石燃料動力。這種趨勢可能會導致每年飆升的家庭力量費用和數百萬噸的新排放。化石燃料在數據中心周圍使用的濃度甚至會帶來健康風險。根據彭博社的一份報告，在數據中心附近的地區，五年來批發電力成本飆升了267％。

Decentralization: A Sustainable Path Forward?

權力下放：前進的可持續道路？

So, what's the solution? Osuri believes decentralization holds the key. Instead of packing chips and energy into mega-data centers, he proposes distributed training across networks of smaller, mixed GPUs. This network could range from high-end enterprise chips to gaming cards in home PCs, unlocking efficiency and sustainability.

那麼，解決方案是什麼？ Osuri認為權力下放是關鍵。他沒有將芯片和能量包裝到大型數據中心，而是提議在較小的混合GPU網絡上分發培訓。該網絡的範圍從高端企業芯片到家用PC中的遊戲卡，釋放效率和可持續性。

Think of it like the early days of Bitcoin mining, where everyday users contributed their processing power and earned rewards. In this case, the "mining" would be training AI models. This approach would democratize AI, giving everyday people a stake in the future while lowering costs for developers.

將其視為比特幣採礦的早期，在這裡，日常用戶貢獻了他們的處理能力並獲得了獎勵。在這種情況下，“採礦”將是訓練AI模型。這種方法將使人工智能民主化，使日常的人們在將來享有股份，同時降低開發人員的成本。

Challenges on the Horizon

挑戰即將到來

Of course, it's not all sunshine and rainbows. Training large-scale models across a diverse array of GPUs requires technological breakthroughs in software and coordination. As Osuri notes, several companies have begun demonstrating aspects of distributed training, but no one has yet pieced it all together and run a complete model.

當然，這並不是所有的陽光和彩虹。培訓各種GPU的大規模模型需要軟件和協調方面的技術突破。正如Osuri所指出的那樣，幾家公司已經開始展示分佈式培訓的各個方面，但是還沒有人將它們拼湊在一起並運行完整的模型。

Incentives are another hurdle. Why would someone volunteer their computer for AI training? What's in it for them? Cracking the incentive code is crucial for making this vision a reality.

激勵措施是另一個障礙。為什麼有人會誌願他們的計算機進行AI培訓？他們中有什麼？破解激勵代碼對於使這一願景成為現實至關重要。

A Necessary Shift

必要的轉變

Despite these challenges, Osuri insists that decentralized AI training is not just a good idea, it's a necessity. By spreading workloads across global networks, we can ease pressure on energy grids, cut carbon emissions, and create a more sustainable AI economy.

儘管面臨這些挑戰，但奧里裡堅持認為，分散的AI培訓不僅是一個好主意，而且是必需的。通過在全球網絡中擴散工作量，我們可以減輕能源網格的壓力，減少碳排放並創造更可持續的AI經濟。

The Future is Decentralized (Maybe)

未來是分散的（也許）

The idea of a decentralized AI training network is intriguing. Imagine earning tokens while your gaming PC helps train the next generation of AI. It's a bold vision. While challenges remain, the potential benefits for the environment and the democratization of AI are hard to ignore. It might sound like a pipe dream now, but hey, who knows? Maybe your trusty gaming rig will be fueling the AI revolution sooner than you think!

分散的AI培訓網絡的想法很有趣。想像一下，當您的遊戲PC幫助培訓下一代AI時，賺取令牌。這是一個大膽的願景。儘管仍然存在挑戰，但很難忽略環境的潛在利益和民主化。現在聽起來可能像是一個白日夢，但是，嘿，誰知道？也許您值得信賴的遊戲鑽機會比您想像的要早日加劇AI革命！