strong ai Meta Stocks Rise as Q1 Ad Sales Maintain Strong, 2025 AI Spending Outlook Lifted

用户投稿 136 0

Meta Stocks Rise as Q1 Ad Sales Maintain Strong, 2025 AI Spending Outlook Lifted

TMTPOST -- Meta Platforms, Inc. shares rose as much as 8% and closed 4.2% higher Thursday. Share rallied after the Facebook operator flexed muscles in digital advertising for last quarter and continued to ratchet up artificial intelligence (AI) spending.

strong ai Meta Stocks Rise as Q1 Ad Sales Maintain Strong, 2025 AI Spending Outlook Lifted

Credit:Pixabay

strong ai Meta Stocks Rise as Q1 Ad Sales Maintain Strong, 2025 AI Spending Outlook Lifted

Meta beat Wall Street expectation both top and bottom line for the quarter ended March 31. Revenue popped 16% year-over-year (YoY) to $42.31 billion, better than analysts estimated $41.38 billion. Though sales cooled compared with a 21% YoY increase for the previous quarter. Diluted earnings per share (EPS) was $6.43 with a 37% YoY increase, compared with projection of $5.25 per share. Operating income soared 35% YoY to $17.56 billion and operating margin increased 3 percentage points YoY to 41%, also above estimated income of $15.52 billion with a margin of 37.5%.

The closed-watched bread & butter advertising earned $41.39 billion with a 16% YoY rise, versus analysts estimated $40.55 billion. Meta needs its advertising business to continue growing in order to fund an expensive expansion in artificial intelligence, which is driving the future of the business through improvements to ads, algorithms and personalization.

Family of Apps revenue added 16% YoY to $41.90 billion for the first quarter, compared with expected $40.89 billion. Revenue from Reality Labs, the hardware unit that houses metaverse technologies, dropped 6.4% YoY to $412 million, missing expectation of $496 million.

"We've had a strong start to an important year, our community continues to grow and our business is performing very well," said Meta founder and CEO Mark Zuckerberg. "We're making good progress on AI glasses and Meta AI, which now has almost 1 billion monthly actives." “I think we’re well positioned to navigate the macroeconomic uncertainty,” Zuckberg said on an earnings conference.

Meta’s guidance was also solid. It expected revenue for the second quarter to be in the range of $42.5 billion to $45.5 billion, in line with the estimate of $44.1 billion.For the full year, Meta called for total expenses to be in the range of $113billion to $118 billion, down from the prior outlook of $114billion to $119 billion.

But perhaps most important, at a time when investors were freaking out about sliding AI spending, Meta boosted its capital expenditure (Capex) forecast. It now sees Capex for this year be in the range of $64billion to $72 billion, increased from its prior outlook of $60billion to $65 billion. That represented a lift of more than 7%.

“This updated outlook reflects additional data center investments to support our artificial intelligence efforts as well as an increase in the expected cost of infrastructure hardware,” Meta said in the earnings release.

Meta Chief Finance Officer Susan Li said the hike in expected infrastructure hardware is the result of “suppliers who source from countries around the world.” The CFO told analysts “there’s just a lot of uncertainty around this, given the ongoing trade discussions”, adding that the company is "working on our end on mitigations by optimizing our supply chain."

Meanwhile, Microsoft Corporation said it plans to ramp up spending the current quarter when releasing its third fiscal quarter ended March 31. The software titan is on pace for Capex of more than $80 billion in its fiscal 2025 year. CFO Amy Hood confirmed the spending will grow at a slower pace later this year. Capex “will grow at a lower rate than FY 2025 and will include a greater mix of short-lived assets, which are more directly correlated to revenue than long-lived assets,” she said.

Google parent Alphabet Inc. reaffirmed its ambitious AI buildout plans with $75 billion Capex guidance for the year. These tech giants’ spending plan seem not to defy skeptics who have predicted a pullback.

"The combo of Microsoft saying demand exceeds supply for next quarter or two and Meta boosting capex by 10%-plus vs. prior to me says cloud capex spending next year is not going down," Mizuho Securities analyst Jordan Klein said in a note. "It will grow again and creates a pretty good setup for Nvidia, Broadcom , TSMC first and foremost."

两位图灵奖得主:自我监督学习是使AI达到人类智能的关键

未来,想要用深度学习的方法来构建达到人类智能水平的人工智能,最关键的一点是什么?

关于这个问题,两位人工智能领域里的大佬达成了一致。在2020年的 ICLR 大会上,两位深度学习领域里的明星科学家,同时也是图灵奖的获得者Yoshua Bengio(约书亚·本吉奥)和Yann LeCun(杨立昆)在谈到深度学习领域里的研究趋势时都表示,自我监督学习是让人工智能推理更像人类的关键。

ICLR,又称国际表征学习大会( International Conference on Learning Representations),是深度学习领域的顶级会议。今年受疫情影响,第一次以线上形式召开。

Yoshua Bengio是人工智能自然语言处理领域的先锋。现为加拿大蒙特利尔大学(University of Montreal)计算机科学与运算系任教授。Yann LeCun常被认为是“卷积网络之父”,目前担任Facebook首席人工智能科学家。

自我监督学习

在深度学习领域,监督学习需要在标记好的数据集上训练AI模型。实际上,由于有用数据的可用性有限,以及处理该数据的计算能力不足,深度学习直到几年前才成为领先的AI技术。而自我监督学习可以做到不再依赖标记数据,而是通过学习数据之间的关系来生成标签,进行训练。杨立昆认为,随着自我监督学习的广泛使用,它将扮演越来越重要的角色。这一步骤也被认为是实现人类智能的关键。

“人类和动物是通过自我监督的模式获得大多数知识的,而不是强化模式。自我监督学习主要是观察世界,并与之互动。这种观察是自发的,而不是在测试条件下完成的。但我们不知道如何让机器获得这样的学习能力。”杨立昆说。

自我监督学习是一种机器学习的“理想状态”,但重点在于如何让机器自动生产数据标签。其中最大的阻碍来自不确定性。通常来说,数据是这样的,它将变量所有可能的值与其出现的概率联系起来。在变量是离散的情况下,它们可以很好地表示不确定性。但目前,研究人员尚未找到一种方法来有效地表示连续变量的分布。

杨立昆指出,解决数据连续分布问题的一种方法是基于能量模型(EBM),该模型学习数据集的数学元素并尝试生成相似的数据集。从历史上看,这种形式的生成建模很难在实际中有所应用,但最近的研究表明,可以对其进行调整以适应复杂的拓扑结构。

Bengio在研讨会上认为,人工智能可以从神经科学领域获得帮助,尤其是对意识和意识加工的探索。他预测,未来的研究会阐明高级语义变量方式与大脑如何处理信息(包括视觉信息)有关。这些变量是人类使用语言交流的事物,它们可能会导致新一代的深度学习模型。

“通过与基础语言学习相结合,我们可以取得很多进展,归根结底我们都在打造能够理解这个世界的模型,以及高层次的概念是如何相互关联的。这是一种联合式的分布。”Bengio 说:“人类的意识处理过程,利用的是有关世界如何变化的假设,这些假设可以理解成为一种更高层次的表达方式。简单来讲,就是我们看到世界的变化,然后想到一句话来解释这种变化。”

没有通用人工智能

除了数据分布的问题外,杨立昆认为缺失的背景知识也是人工智能无法达到人类智能水平的阻碍之一。比如,大多数人可以在30小时内学会驾驶汽车,因为他们已经掌握了有关汽车行为的物理模型。相比之下,自动驾驶汽车上部署的强化学习模型从零开始,他们必须先犯成千上万个错误,才能确定哪些操作是无害的。

“显然,我们需要能够构建模型来学习世界,这就是进行自监督学习的原因——运行预测世界模型,能够让系统进行更快地学习。从概念上讲,这相当简单,除非是在那些我们无法完全预测的不确定环境中。”杨立昆说。同时,他认为,想要实现通用人工智能(AGI),就算有自监督学习和神经科学学习的加持,也是不够的。

通用人工智能是指机器获得人类水平的智能。一些研究人员将通用人工智能称为强AI(strong AI)或者完全AI(full AI),或称机器具有执行通用智能行为(general intelligent action)的能力。

杨立昆表示,这是因为智力,尤其是人类的智力,是非常特殊的。“AGI是不存在的,根本没有通用人工智能这回事。我们可以谈论老鼠水平的智力、猫的智力、狗的智力,或者人类的智力,但还根本谈不上通用人工智能。”杨立昆说。

但是Bengio相信,机器最终将能够在无需体验的情况下获得关于世界的各种知识的能力,这很可能是而是通过习得可以语言化的知识来实现。

“我认为这也是人类的一个巨大优势,相比于其他动物来说,人类之所以聪明,是因为我们有我们自己的文化,让我们能够解决这个世界的问题。要想让人工智能在现实世界中发挥作用,我们需要它不仅仅是能够翻译的机器,而是能够真正理解自然语言的机器。”Bengio说。

相关问答

onlythestrongsurvive艾弗森官方版重置版APPv8.6.13-图吧地图

根本上还是市场对高估值AI概念股信心的脆弱导致了大跌5、国庆节特别策划:我的名字叫“国庆”国庆日围绕探索走出一条体现政治要求、满足党员需求、提高教育实效...

曼彻斯特大学(英国)PhD直招有多难?有没有2024最新计算机科...

[回答]很高兴能为你解答关于曼彻斯特大学(英国)PhD直招的相关问题。曼彻斯特大学(英国)在全球计算机科学领域极具影响力,其PhD项目自然吸引着众多优秀...

考文垂大学PhD直招有多难?有没有2024最新信息系统学专业相...

[回答]们详细了解一下该项目的直招情况。考文垂大学PhD直招的申请时间通常在每年1月。需要提醒的是,由于申请竞争激烈,建议尽早提交申请材料。...考文...

1.Longlongago,_______(ðeəz)astrong_______('d&...

[最佳回答]1.there'sgiantcastle2.engineer3.C4.C5.flew6.Doeshave7.dodoyour

爱丁堡大学PhD直招有多难?有没有2024最新计算机工程专业相...

[回答]们详细了解一下该项目的直招情况。爱丁堡大学PhD直招的申请时间通常在每年1月。需要提醒的是,由于申请竞争激烈,建议尽早提交申请材料。...爱丁...

koznak2022(中国)IOS/手机APP官网下载V3.71.32;

saidonTwitterthattheactorshad“harassed”Mr.Pence,andheissuedabattlecrytohissupporte...

奥克兰大学(新西兰)PhD直招有多难?有没有2024最新计算机科...

[回答]很高兴能为你解答关于奥克兰大学(新西兰)PhD直招的相关问题。奥克兰大学(新西兰)在全球计算机科学领域极具影响力,其PhD项目自然吸引着众多优秀...

爱丁堡大学PhD直招有多难?有没有2024最新法学专业相关的直...

[回答]详细了解一下该项目的直招情况。爱丁堡大学PhD直招的申请时间通常在每年1月。需要提醒的是,由于申请竞争激烈,建议尽早提交申请材料。...爱丁堡...

奥克兰大学(新西兰)PhD直招有多难?有没有2024最新数学专业...

[回答]很高兴能为你解答关于奥克兰大学(新西兰)PhD直招的相关问题。奥克兰大学(新西兰)在全球数学领域极具影响力,其PhD项目自然吸引着众多优秀学子的...

巴斯大学PhD直招有多难?有没有2024最新软件工程专业相关的...

[回答]详细了解一下该项目的直招情况。巴斯大学PhD直招的申请时间通常在每年1月。需要提醒的是,由于申请竞争激烈,建议尽早提交申请材料。以...巴斯大...

抱歉,评论功能暂时关闭!