AI培训的能源危机：与格雷格·奥苏里（Greg Osuri）的权力下放能否提供解决方案？

人工智能培训要求巨大的能量。格雷格·奥苏里（Greg Osuri）提出权力下放，以解决缓解能源危机的解决方案。

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繁荣在这里，但是它有一个肮脏的小秘密：它像加油站一样笨拙的能量。随着AI模型变得更大和脑，训练它们所需的能量正在飙升，这引起了人们对即将发生的能源危机的担忧。 Akash Network的创始人Greg Osuri正在敲响警钟，这表明该行业低估了随之而来的计算需求和环境成本。

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，数据中心是AI背后的强力，已经消耗了数百兆瓦的化石燃料动力。这种趋势可能会导致每年飙升的家庭力量费用和数百万吨的新排放。化石燃料在数据中心周围使用的浓度甚至会带来健康风险。根据彭博社的一份报告，在数据中心附近的地区，五年来批发电力成本飙升了267％。

Decentralization: A Sustainable Path Forward?

权力下放：前进的可持续道路？

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.

那么，解决方案是什么？ Osuri认为权力下放是关键。他没有将芯片和能量包装到大型数据中心，而是提议在较小的混合GPU网络上分发培训。该网络的范围从高端企业芯片到家用PC中的游戏卡，释放效率和可持续性。

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.

将其视为比特币采矿的早期，在这里，日常用户贡献了他们的处理能力并获得了奖励。在这种情况下，“采矿”将是训练AI模型。这种方法将使人工智能民主化，使日常的人们在将来享有股份，同时降低开发人员的成本。

Challenges on the Horizon

挑战即将到来

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.

当然，这并不是所有的阳光和彩虹。培训各种GPU的大规模模型需要软件和协调方面的技术突破。正如Osuri所指出的那样，几家公司已经开始展示分布式培训的各个方面，但是还没有人将它们拼凑在一起并运行完整的模型。

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.

激励措施是另一个障碍。为什么有人会志愿他们的计算机进行AI培训？他们中有什么？破解激励代码对于使这一愿景成为现实至关重要。

A Necessary Shift

必要的转变

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.

尽管面临这些挑战，但奥里里坚持认为，分散的AI培训不仅是一个好主意，而且是必需的。通过在全球网络中扩散工作量，我们可以减轻能源网格的压力，减少碳排放并创造更可持续的AI经济。

The Future is Decentralized (Maybe)

未来是分散的（也许）

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培训网络的想法很有趣。想象一下，当您的游戏PC帮助培训下一代AI时，赚取令牌。这是一个大胆的愿景。尽管仍然存在挑战，但很难忽略环境的潜在利益和民主化。现在听起来可能像是一个白日梦，但是，嘿，谁知道？也许您值得信赖的游戏钻机会比您想象的要早日加剧AI革命！