All’s well that’s Blackwell
凡事皆大吉祥
Nvidia’s boss dismisses fears that AI has hit a wall
英伟达老板驳斥人工智能已碰壁的担忧
But it’s “urgent” to get to the next level, Jensen Huang tells The Economist
黄仁勋告诉《经济学人》：突破下一个高度已是迫在眉睫
WHEN SAM ALTMAN, boss of OpenAI, posted a gnomic tweet this month saying “There is no wall,” his followers on X, a social-media site, had a blast. “Trump will build it,” said one. “No paywall for ChatGPT?” quipped another. It has since morphed from an in-joke among nerds into a serious business matter.
当 OpenAI 首席执行官山姆·奥尔特曼（Sam Altman）本月在 X（社交媒体平台）发布一条意味深长的推文，称"没有墙"时，他的追随者们纷纷调侃。有人说"特朗普会修墙"，另一些人则揶揄"ChatGPT没有付费墙吗？"这个玩笑已从技术圈的冷幽默演变成一个严肃的商业话题。
The wall in question refers to the view that the forces underlying improvements in generative artificial intelligence (AI) over the past 15 years have reached a limit. Those forces are known as scaling laws. “There’s a lot of debate: have we hit the wall with scaling laws?” Satya Nadella, Microsoft’s boss, asked at his firm’s annual conference on November 19th. A day later Jensen Huang, boss of Nvidia, the world’s most valuable company, said no.
所谓的"墙"，指的是过去15年里推动生成式人工智能（AI）进步的底层力量已经达到极限的观点。这些力量被称为"缩放定律"。"关于我们是否已经触及缩放定律的极限，存在大量争议。"微软CEO 萨提亚·纳德拉（Satya Nadella） 在11月19日的公司年会上如是说。次日，全球市值最高公司英伟达CEO黄仁勋则明确予以否定。
Scaling laws are not physical laws. Like Moore’s law, the observation that processing performance for semiconductors doubles roughly every two years, they reflect the perception that AI performance in recent years has doubled every six months or so. The main reason for that progress has been the increase in the computing power that is used to train large language models (LLMs). No company’s fortunes are more intertwined with scaling laws than Nvidia, whose graphics processing units (GPUs) provide almost all of that computational oomph.
缩放定律并非物理定律。类似摩尔定律观察到半导体处理性能大约每两年翻一番，缩放定律反映了近年来AI性能大约每六个月翻一番的趋势。推动这一进步的主要原因，是用于训练大语言模型（LLMs）的计算力持续增长。没有哪家公司的命运与缩放定律的关系比英伟达更紧密——其图形处理器（GPUs）几乎提供了所有这些计算能量。
On November 20th, during Nvidia’s results presentation, Mr Huang defended scaling laws. He also told The Economist that the first task of Nvidia’s newest class of GPUs, known as Blackwells, would be to train a new, more powerful generation of models. “It’s so urgent for all these foundation-model-makers to race to the next level,” he says.
在11月20日英伟达业绩发布会上，黄仁勋为缩放定律进行了辩护。他还告诉《经济学人》，英伟达最新一代 Blackwell GPU 的首要任务，将是训练一个更强大的新一代模型。"所有基础模型制造商加速突破到下一个层级，这是极其紧迫的。"他说。
The results for Nvidia’s quarter ending in October reinforced the sense of upward momentum. Although the pace of growth has slowed somewhat, its revenue exceeded $35bn, up by a still-blistering 94%, year on year (see chart). And Nvidia projected another $37.5bn in revenues for this quarter, above Wall Street’s expectations. It said the upward revision was partly because it expected demand for Blackwell GPUs to be higher than it had previously thought. Mr Huang predicted 100,000 Blackwells would be swiftly put to work training and running the next generation of LLMs.
英伟达截至10月的季度业绩，进一步巩固了这种持续上升的势头。尽管增长速度略有放缓，但营收仍超过350亿美元，同比增长94%。公司预计下一季度营收将达375亿美元，超出华尔街预期。公司表示，这一上调部分源于预期 Blackwell GPU 的需求将高于此前预计。黄仁勋预测，10万台 Blackwell 将迅速投入训练和运行下一代大语言模型。
Not everyone shares his optimism. Scaling-law sceptics note that OpenAI has not yet produced a new general-purpose model to replace GPT-4, which has underpinned ChatGPT since March 2023. They say Google’s Gemini is underwhelming given the money it has spent on it.
并非所有人都对此持乐观态度。质疑缩放定律的人士指出，OpenAI尚未推出替代GPT-4的通用新模型，自2023年3月以来一直使用该模型支持ChatGPT。他们还认为，谷歌的Gemini模型与其投入的资金不成正比。
But, as Mr Huang notes, scaling laws not only apply to the initial training of LLMs, but also to the use of the model, or inference, especially when complex reasoning tasks are involved. To explain why, he points to OpenAI’s latest model, o1, which has stronger reasoning capabilities than GPT-4. It can do advanced maths and other complex tasks by taking a step-by-step approach that its maker calls “thinking”. This enhanced inference process uses far more computing power than a typical ChatGPT response, Mr Huang says. “We know that we need more compute whatever the approach is,” he says.
但正如黄仁勋指出的，缩放定律不仅适用于大语言模型的初始训练，还适用于模型的使用，尤其是在涉及复杂推理任务时。为解释这一点，他提到了OpenAI最新的o1模型，该模型的推理能力强于GPT-4。通过逐步思考的方法，它可以完成高级数学和其他复杂任务。黄仁勋表示，这种增强的推理过程比典型的ChatGPT响应消耗更多计算资源。"无论采用何种方法，我们都知道需要更多计算能力。"他说。
The more AI is adopted, the more important inference will become. Mr Huang says that Nvidia’s previous generations of GPUs can be used for inference, but that Blackwells will make performance dozens of times better. Already at least half of Nvidia’s infrastructure is used for inference.
随着AI被越来越广泛地采用，推理过程将变得更加重要。黄仁勋表示，英伟达之前的GPU一代可用于推理，但 Blackwell 将使性能提高数十倍。目前，英伟达至少一半的基础设施已用于推理。
Mr Huang clearly has an interest in portraying scaling in the best possible light. Some sceptics question how meaningful the advances in reasoning are. Although a handful of business models are being disrupted, many firms are struggling to adopt AI at scale, which could eventually weigh on demand for the technology. These are early days, though. Tech giants continue to spend big on GPUs, and Mr Huang points out that new technologies take time to digest. Nvidia’s back is not against the wall yet.
黄仁勋显然有动机以最佳角度描绘技术缩放。一些怀疑论者质疑推理能力的进步有多么显著。尽管少数商业模式已经被颠覆，但许多公司在大规模采用AI方面仍面临困难，这可能最终会抑制对该技术的需求。不过，现在毕竟是技术发展的早期阶段。科技巨头们继续大举投资GPU，黄仁勋指出，新技术需要时间消化。英伟达目前尚未陷入困境。
重难点词汇
hit a/the wall 达到极限
gnomic /ˈnəʊmɪk/ adj. ，格言警句式的，精辟的，深奥的
quip /kwɪp/ n. 俏皮话，妙语 v. 讲俏皮话；讥讽；嘲弄；打趣
morph /mɔːf/ v. (尤指很大的、质的)改变; 变化
in-joke n. 圈子里的笑话；行内笑话
perception /pəˈsepʃn/ n. 知觉，感知 ；洞察力，悟性；看法，见解
intertwine /ˌɪntəˈtwaɪn/ v. 交织，缠绕；紧密相连
oomph /ʊmf/ n. 精力，特质，气质
blistering /ˈblɪstərɪŋ/ adj. 描述体育动作）迅速的，劲头十足的；酷热的；言辞激烈的，尖刻的，愤怒的
underpin /ˌʌndəˈpɪn/ v. 加强，巩固，构成（…的基础等）
underwhelming /ˌʌndəˈwɛlmɪŋ/ adj. 令人失望的，平庸的，索然无味的
portray /pɔːˈtreɪ/ v. 描绘，描写；将…描写成，给人以某种印象，表现；扮演（某角色）
weigh on 加重…的思想负担；使焦虑不安；使担忧
感谢大家阅读，Enjoy!
来源：The Economist- November 23-29, 2024
原文标题：Nvidia’s boss dismisses fears that AI has hit a wall
AI翻译，译文仅供参考
往期阅读
外媒怎么报道李子柒回归？| 外刊阅读
「经济学人」马斯克变了！Elon Musk’s transformation, in his own words | 外刊阅读
介绍一本金句频出、提升认知的英语原版书
崔娃新书 Into the Uncut Grass | 英文畅销书