Facebook Earnings, Generative AI and Messaging Monetization, Open Source and AI

This update about Facebook earnings, Generative AI and messaging monetization, and open source AI was published on Wednesday, May 3rd, 2023. Good morning. On Monday sharp tech we discuss why I have been and continue to be skeptical of micro-payments, including Twitter's plans in this area. Meanwhile, sharp China discuss Ambassador Lu Shai-E's in Cinderio Marks and the EU, while greatest of all talk continues to be the go-to source for coverage of the ongoing MBA playoffs. You can listen to these podcasts at the links above, or add them to your podcast player using the links in your show notes.

这则更新是关于Facebook的盈利、生成式人工智能和消息驱动的货币化以及开源人工智能的。它于2023年5月3日星期三发布。早上好！在周一的“尖端技术”上，我们讨论了为什么我对微支付持怀疑态度，包括Twitter在此领域的计划。同时，“尖端中国”讨论了卢沙埃大使在辛德里奥马克斯和欧盟的事情，而“最伟大的话题”则继续成为覆盖持续进行的MBA季后赛的主要来源。你可以点击以上链接听这些播客，或将它们添加到你的播客播放器中，使用你的节目笔记中的链接。

On to the update. Facebook earnings from the Wall Street Journal. Facebook parent Meta Platforms Inc on Wednesday reported its first increase in sales in nearly a year due to continued improvements in its advertising business, as the company continues to pair back spending during what chief executive Mark Zuckerberg has called a, quote, year of efficiency, end quote. Mr Zuckerberg attributed some of these gains to Reels, the company's short form video product.

接下来，我们来谈一下关于Facebook收益的更新消息。根据华尔街日报的报道，Facebook的母公司Meta Platforms公司周三报告称，由于公司不断改进其广告业务，并在首席执行官马克·扎克伯格所说的“效率年”期间削减支出，公司的销售额近一年来首次增长。扎克伯格认为其中部分收益归功于Reels，即公司的短视频产品。

Mr Zuckerberg said that Reels is increasing overall app engagement and that the company believes it is gaining share in the short form video market. The company reported revenue of $28.6 billion, up 3% from a year prior, and ahead of expectations of nearly 27.7 billion, according to analysts surveyed by FACSET. That snapped a streak of three quarters in which Meta's revenue had retreated from the year prior. The only time that has occurred since the company went public in 2012. Share surged by more than 12% in after hours trading, as the company also forecast that second quarter revenue could reach as high as $32 billion.

扎克伯格先生表示，Reels在增加整体应用程序的参与度，并且公司认为正在赢得短视频市场的份额。根据FACSET调查的分析师的预期，该公司报告的收入为286亿美元，比前一年增长3％，并且超过近277亿美元的预期。这打破了自公司2012年上市以来连续三个季度收入下降的记录。公司股票在盘后交易中大涨超过12％，因为该公司还预测第二季度的收入可能达到320亿美元。

I know I'm a bit late in getting to these earnings, but frankly, I already covered them six months ago and met a myth to review. Myth one, users are deserting Facebook. In fact, Meta increased users sequentially in every single market, cresting the 3 billion daily active user mark. I will note that CFOs recently dodged a question asking about overall engagement on the Facebook app specifically. Myth two, Instagram engagement is plummeting. In fact, Instagram engagement continues to surge thanks to increased Reels engagement.

我知道我有点迟到了，因为这些财报已经公布了六个月了，但坦白说，我已经在六个月前就对它们进行了评论并提出了一个谣言要进行审核。谣言一：用户正在离开 Facebook。事实上，Meta 在每个市场中都有用户数量连续增长，超过了每天活跃用户达到 30 亿。我要指出的是，CFO 最近回避了一个关于 Facebook 应用程序整体参与度的问题。谣言二：Instagram 的参与度正在暴跌。事实上，由于 Reels 参与度的增加，Instagram 的参与度仍在不断增长。

Notably, it is this increased engagement that drives Meta's optimism that Reels will be revenue neutral by the end of the year. Yes, they take time away from feed and stories, which monetize more efficiently because users sweat through them more quickly, but that is how much more time users are spending with Reels. Myth three, TikTok is dominating. This increased Reels engagement is not happening in a vacuum. Meta believes they are taking share and all the evidence I have seen is that that is indeed the case. Myth four, advertising is dying. Obviously not.

值得注意的是，正是这种增加的使用率驱动着Meta的乐观，认为Reels将在年底实现收入平衡。是的，Reels会从Feed和Stories中抢走用户时间，而这些内容更具有效的变现能力，因为用户会更快地完成它们。但用户花费在Reels上的时间却更多。第三个谣言是TikTok正在主导市场。这种增加的Reels使用率并非孤立发生的。Meta认为他们正在占据市场份额，而我所见到的所有证据都表明，这的确是事实。第四个谣言是广告即将消亡。显然不是这样的。

In fact, there is a case to be made that Meta had the single most remarkable advertising quarter in a very long time. The below is a chart that I have been referencing in Facebook's earnings updates for years. See chart about ad, daily active user, price per ad and ad impression growth rates over time. To briefly review why I pay attention to the graph, there are three ways to increase ad revenue. One, increase the number of users. Two, increase the number of ads seen by each user. Three, increase the price that advertisers pay per ad.

实际上，我们可以说Meta在最近的一季度中拥有了最为不同寻常的广告表现。以下是一个图表，这个图表我在Facebook的财报更新中已经引用了多年。可以看到图表中有关广告、活跃用户、广告价格和广告印象增长率的信息。简单回顾一下我为什么要注意这个图表，主要有三个增加广告收入的方法：一，增加用户数量；二，增加每个用户看到的广告数量；三，提高广告主每个广告的支付价格。

The number of users doesn't really move the needle for men anymore, at least compared to a decade ago. Although, as noted above, the number of users does continue to increase. I'll be at mostly in lower revenue reasons of the world. The number of ad scene preezer can increase artificially or organically. An artificial increase would be increasing ad load. Meta stopped doing that back in 2017, which is why impressions growth started dropping. An organic increase, on the other hand, can come from either increased engagement, thus more opportunities to show ads, and or through increased surface area. An example of the latter is stories, which re-excelerated impression growth in 2018.

相比10年前，用户数量对男性来说并不再有太大影响了。尽管如上所述，用户数量仍然在增加。我将主要在世界上营收较低的地区。广告展示次数的增加可以是人为的或有机的。人为增加可以是增加广告负荷。Meta在2017年停止了这样做，这就是印象增长开始下降的原因。另一方面，有机增长可以通过增加参与度从而提供更多展示广告的机会，或通过增加表面积。后者的一个例子是故事，这在2018年重新加速了印象增长。

The price per ad, which remember is set by an auction, not meta itself, can increase in two ways as well. First, medicine can simply do a better job of targeting ads, increasing the return on advertising spend, row-ass for advertisers, increasing their willingness to pay. However, because these prices are set at auction, the much more significant driver of price are the number of impressions available. That is why every time impressions increase, such as when stories took off, for example, or with the recent increase in reels, the price per ad decreases and vice versa.

每个广告的价格是由拍卖设置的，而不是由 Meta 自己设置的。同时，广告价格也有两种增长方式。第一种是医药公司可以更精准地定位广告，提高广告开支的回报率，从而提高广告商的报价意愿。然而，由于这些价格是通过拍卖设置的，更显著的价格驱动因素是 impressions 的数量。这就是为什么每当 impressions 增加，例如像 stories 的兴起或者最近 reels 的增加，每个广告的价格就会降低，反之亦然。

Note though, that that connection was significantly loosened last quarter. For the first time in years, the sequential change in impressions growth, from 23% to 26%, and the sequential change in price per ad growth, from negative 22% to negative 17%, moved in the same direction. In other words, meta increased impressions and what prices per ad fell, they fell by less than you would expect given the impressions growth. Here's the part of the earnings call where CFO Susan Lee mentioned the percentages. In Q1, the total number of ad impressions served across our services increased 26%, and the average price per ad decreased 17%. Impression growth was primarily driven by Asia Pacific and rest of world. The year over year decline in pricing was primarily driven by strong impression growth, especially from lower monetizing services and regions, foreign currency depreciation, and lower advertising demand. While overall pricing remains under pressure from these factors, we believe our ongoing improvements to add targeting and measurement are continuing to drive improved results for advertisers.

请注意，上季度这种联系已经显着减弱。多年来，印象增长的连续变化从23%增加到26%，每广告价格增长的连续变化从负22%降至负17%，意味着Meta增加了印象，广告价格降低了，但降价幅度比你预计的要小。以下是首席财务官Susan Lee在收益电话中提到的百分比。在第一季度，我们服务中提供的广告印象总数增长了26%，每广告平均价值下降了17%。印象增长主要受亚太地区和世界其他地区的推动。年度定价下降主要由于强劲的印象增长，特别是来自较低货币化的服务和地区，货币贬值和广告需求减少。尽管整体定价仍受到这些因素的压力，但我们相信我们持续改进广告定位和测量的努力将继续推动广告主的改善结果。

Meta has been well known for sandbagging the strength of their business for years. The company is very good at dumping all bad news in one call, giving a disappointing forecast, and then beating that forecast for the next several years. This commentary is a part of that rich tradition. Lee doesn't call this divergence in derivatives at all, and none of the analysts on the call asked her about it. That last sentence, though, almost certainly explains what is going on. Meta has cracked the post-ATT code when it comes to ad targeting.

Meta公司多年来一直以低估业务实力而闻名。该公司非常善于在一个电话会议中披露所有坏消息，给出令人失望的预测，然后在接下来的几年中超越这个预测。这篇评论也是这种丰富传统的一部分。李女士根本没有提到衍生品的分化，电话会议上也没有分析师问过她。然而，最后一句话几乎可以解释发生了什么。Meta公司在广告定位方面已经破解了突破ATT（苹果公司推出的隐私保护标准）的代码。

Myth 5, Meta spending is a waste. C.O. Mark Zuckerberg said that Meta was monetizing better because of AI, the buildout of which has been the biggest factor in Meta's increased capital expenditure. Our AI work is also improving monetization. Real monetization efficiency is up over 30% on Instagram and over 40% on Facebook quarter over quarter. Daily revenue from advantage plus shopping campaigns is up 7x in the last six months. Needless to say, that spending was not a waste. Indeed, I argued in that article that it would make Meta even stronger. Quote, that last point is perhaps the most important.

神话5：“元花费”是浪费。Facebook首席运营官马克·扎克伯格表示，由于人工智能的发展，Meta的盈利能力得到了提高，这也是Meta增加资本支出的最大因素。我们的人工智能研究也正在提高盈利能力。在Instagram上，实现了超过30%的真正盈利效率提高，而在Facebook上则超过了40%。在过去的六个月中，优势加购物广告的日收入翻了7倍。毋庸置疑，这些支出并不是浪费。事实上，我在那篇文章中强调了这一点，认为这将使Meta更加强大。引用一句话，这也许是最重要的一点。

ATT hurt Meta more than any other company because it already had by far the largest and most finely tuned ad business. But in the long run, it should deepen Meta's moat. This level of investment simply isn't viable for a company like Snap or Twitter or any of the other also brands in digital advertising. Even beyond the fact that Snap relies on cloud providers instead of its own data centers.

ATT对Meta的伤害比任何其他公司都要大，因为它已经拥有迄今为止最大、最精细调整的广告业务。但从长远看，这应该加深Meta的壕沟。这种投资水平对像Snap或Twitter这样的公司或其他数字广告品牌来说根本不可行。甚至超出Snap依赖云提供商而不是自己的数据中心这个事实之外。

When you combine the fact that Meta's ad targeting will likely start to pull away from the field outside of Google with the massive increase in inventory that comes from Reels, which reduces prices, it will be a wonder why any advertiser would bother going anywhere else. End quote. It's worth knowing that Meta's revenue increased even as YouTube's decreased. Given that YouTube is the part of Google's business as the most exposed to ATT, this shift in fortune seems notable. Snap, meanwhile, missed on revenue once again, plunging its market cap to $13 billion. Meta took quite the blow from Apple, but it seems clear that not only has the company survived, it very well may be stronger than ever, particularly in terms of its competitive position.

当Meta的广告定位开始从Google的领域中脱颖而出并且Reels的库存大量增加从而降低价格时，让任何广告商选择其他平台都感到惊讶。值得一提的是，即使YouTube的营收下降，Meta的营收却增长了。考虑到YouTube是Google业务中最容易受到ATT影响的部分，这种财富的转移似乎值得注意。Snap却一次又一次地错过了营收，导致其市值降至130亿美元。Meta在运营上受到了苹果的冲击，但显然该公司不仅存活下来了，而且极有可能比以往任何时候都更强大，特别是在竞争地位方面。

This brings up what was, to me, the most shocking part of the earnings call. I don't normally pick on analysts, but this question baffled me. Mark, this one's a bit more philosophical. But if we compare where Meta is today kind of pre-idea-fa cookie deprecation kind of world, how do you see the relative competitive position in digital advertising in light of all of these technical integrations and AI tools? Because I think the prevailing view was that privacy would level the playing field, but it certainly doesn't seem that way. If I've said it once, I've said it a million times. Privacy does not level the playing field. In fact, it works in the exact opposite direction, not unlike the relationship between impressions and price per app. The more privacy restrictions there are, the more of an advantage there is for the most dominant players with the most first party data, and the most capability to work around the restrictions. This was the case in Europe with GDPR. It was the case with ATT, and it bottles my mind that people still have it in their heads that it would turn out any other way. Yes, the biggest players feel the most pain on an absolute basis, but that's because they are big. Their position relative to their competitors ends up being stronger, and that certainly appears to be the case with Meta.

这让我感到最震惊的是收益电话中的一个问题。我通常不会挑剔分析师，但这个问题让我感到困惑。马克，这个问题有点哲学性。但是，如果我们将Meta今天处于“想法饼干去警告”之前的世界与数字广告的相对竞争地位相比较，考虑到所有这些技术集成和AI工具，您如何看待这个问题？因为我认为，普遍的观点是隐私会扭转这个局势，但它似乎并不是这样的。如果我说过一次，我已经说过一百万次。隐私不会让竞争更加公平。事实上，它朝着完全相反的方向发展，就像印象和应用每次价格之间的关系一样。隐私限制越多，最具有优势的玩家也就越具有竞争优势，他们拥有最多的第一方数据和绕过限制的能力。这在欧洲的GDPR案例中是如此，也是在ATT中是这样的，令人难以置信的是，人们仍然认为它会有其他的结果。是的，最大的玩家在绝对基础上感受到更多的痛苦，但那是因为他们很大。相对于竞争对手，他们的位置变得更加强劲，这在Meta的情况中显然是如此。

Generative AI and messaging monetization.

生成式人工智能和消息变现。 这句话指的是利用生成式人工智能技术开发出聊天机器人等工具，并通过这些工具进行消息变现。简单来说，就是通过智能化的方式来赚钱。

Two bits of background here. First, one of the areas of generative AI that I am the most interested in is its application advertising. Instead of having to supply assets that Meta can A-B test, Meta could generate images for endless A-B test against a dizzying array of cohorts, delivering personalized advertising not just on a targeting basis, but on a content basis. Second, last year Meta disclosed that it's quick to messaging advertising business, which connects would-be customers directly with advertisers in WhatsApp and Messenger, was running at a $10 billion run rate. Meta didn't update that number, but he said it continued to grow last quarter.

两点背景信息：首先，我最感兴趣的生成 AI 领域之一是其在广告营销方面的应用。Meta 可以生成无数种不同的图片供广告 A-B 测试，从而为各种族群提供个性化广告，不仅基于目标定向，而且基于内容定向。其次，去年，Meta 披露其快速通讯广告业务在 WhatsApp 和 Messenger 上帮助顾客直接联系广告商，预计年收入达 100 亿美元。Meta 没有更新该数字，但称上个季度业务仍在增长。

What is intriguing is the potential to put these two things together. Zuckerberg said in response to a questionable generative AI's impact on Meta's ad business. I also think that there's going to be a very interesting convergence between some of the AI agents and messaging and business messaging, where right now we see a lot of the places where business messaging is most successful are places where a lot of businesses can afford to basically have people answering a lot of questions for people in engaging them with them and chat. And obviously once you light up the ability for tens of millions of small businesses to have AI agents acting on their behalf, you'll have way more businesses that can afford to have someone engaging in chat with customers I think that that could be a pretty big opportunity to. So in short, this messaging advertising works better if you have well someone to mesh with. Large language models mean that soon enough every business will be able to offer just that.

有趣的是将这两个事物结合起来的潜力。扎克伯格在回答有关生成AI对Meta广告业务影响的问题时表示。我认为，一些AI代理和消息传递以及商业消息之间将会出现非常有趣的融合，目前我们看到商业消息最成功的地方往往是许多企业可以承担基本上让人们回答许多问题并与他们进行交流的人。显然，一旦您启用AI代理以代表数以千万计的小型企业，您将会有更多的企业可以负担雇用人员与客户进行聊天交流，我认为这可能是一个相当大的机会。因此，简而言之，如果您有能够与之融合的人，则此消息广告效果更佳，而大型语言模型意味着很快每家企业都将能够提供这一服务。

Open source and AI.

开源与人工智能。 这是指开放源代码的软件和技术在人工智能领域的应用和发展。开源软件通常可以免费获取和修改，因此开源的AI技术能够激发更多的创新和合作，而不需要额外的许可费用。AI的开源技术也能够降低人工智能的门槛，使更多的人可以参与到这个领域中来。

Zuckerberg emphasized open source and is prepared remarks. Now right now most of the companies that are training large language models have business models that lead them to a closed approach to development. And I think that there is an important opportunity in the industry to help create an open ecosystem. And if we can help be a part of this, then much of the industry I think will standardize on using these open tools and help improve them further. So this will make it easier for other companies to integrate with our products and platforms as we enable more integrations. And that will help our product state the leading edge as well.

扎克伯格强调了开放源代码和准备好的言论。现在大多数正在训练大语言模型的公司都有导致他们采用封闭式开发方法的商业模式。我认为行业中存在一个重要的机会，可以帮助创建一个开放的生态系统。如果我们能够参与其中，那么我认为，许多公司将会采用这些开放的工具，并进一步完善它们。这将使其他公司更容易与我们的产品和平台集成，随着我们推出更多的集成，这也将有助于使我们的产品保持领先地位。

Our approach to AI and our infrastructure has always been fairly open. We open source many of our state-of-the-art models so people can experiment and build with them. This quarter we released our LOM to researchers. It has 65 billion parameters but outperformed larger models and has proven quite popular. We've also open sourced three other ground breaking visual models along with their training data and model weights, segment anything, dyno V2, and our animated Drongs tool. And we've gotten some positive feedback on all of those as well.

我们一直采用相对开放的方法来处理人工智能及其基础设施。我们将我们的最先进模型开源，让人们可以进行实验和构建。本季度，我们向研究人员发布了我们的LOM。虽然它有65亿个参数，但其表现却优于更大的模型，并且受到了很受欢迎。我们还开源了其他三个具有突破性的视觉模型及其训练数据和模型权重，例如segment anything、dyno V2和我们的Drongs动画工具。这些都得到了一些积极的反馈。

LOMA is not truly open sourced. It can't be used for commercial applications. But as I detailed it in an update earlier this year, it's availability led to an explosion and experimentation and innovation, particularly in terms of running LLMs locally. Was Zuckerberg hinting at truly open sourcing some of Meta's models?

LOMA不是真正的开源，不能用于商业应用。但正如我今年早些时候发布的更新所详细介绍的那样，它的可用性导致了爆炸性的实验和创新，特别是在本地运行LLMs方面。马克·扎克伯格暗示Meta的一些模型是否真正开源？

Zuckerberg expanded on the open source point in the question and answer portion. I think that there's an important distinction between the products we offer and a lot of the technical infrastructure, especially the software that we write to support that. And historically, whether it's the open compute project that we've done or just open sourcing a lot of the infrastructure that we've built, we've historically open sourced a lot of that infrastructure, even though the product themselves are obviously, we're not, we have an open source to the code for our core products or anything like that. And the reason why I think why we do this is that unlike some of the other companies in the space, we're not selling a cloud computing service where we try to keep the different software infrastructure that we're building proprietary.

扎克伯格在问答环节中进一步扩展了开源的观点。我认为我们提供的产品和很多技术基础设施之间存在着重要的区别，特别是我们编写的支持软件。历史上，无论是我们进行的开放式计算项目，还是仅仅开源我们构建的大量基础设施，我们历来开源了很多基础设施，尽管产品本身显然不是这样​​的，我们并没有开源我们核心产品的代码或类似的东西。我认为我们这样做的原因在于，与该领域的一些其他公司不同，我们不是在销售云计算服务，我们试图将不同的软件基础架构保持专有。

For us, it's way better if the industry standardizes on the basic tools that we're using. And therefore, we can benefit from the improvements that others make. And others use of those tools, in some cases, like open compute, drive down the costs of those things which make our business more efficient too. So I think to some degree, we're just playing a different game on the infrastructure than companies like Google or Microsoft or Amazon and that creates different incentives for us. Meta has long been very good at pursuing a commoditizer complement strategy in terms of its infrastructure. This is an articulation of exactly that.

对于我们而言，如果行业能够标准化我们所使用的基本工具，那将会更好。这样，我们就能够获益于其他公司所作出的改进。在某些情况下，像开放计算一样的工具的普及，可以降低我们业务中需要的成本，使我们的公司更加高效。因此，我认为，在基础设施上，我们与谷歌、微软或亚马逊等公司正在玩不同的游戏，这为我们创造了不同的激励机制。Meta公司一直在追求基础设施上的组合策略。这正是这一战略的表述。

Meta's values in the network and its apps, it would be counterproductive to seek some sort of differentiation anywhere else in the value chain. Indeed, the best outcome is that the industry ends up working together to drive Meta's cost-stone. Zuckerberg continued. So overall, I think that's going to lead us to do more work in terms of open sourcing some of the lower level models and tools.

在 Meta 公司的网络和应用程序价值中，如果在价值链的其他任何地方寻求某种差异化，反而会适得其反。事实上，最好的结果就是行业合作，共同推动 Meta 公司成本的降低。马克·扎克伯格继续说道。因此，总的来说，我认为这将促使我们更多地开源低级别模型和工具的工作。

But of course, a lot of the product work itself is going to be specific and integrated with the things that we do. So it's not that everything we do is going to be open. Obviously, a bunch of this needs to be developed in a way that creates unique value for our products. But I think in terms of the basic models, I would expect us to be pushing and helping to build out an open ecosystem here, which I think is something that's going to be important. That sure sounds like Meta's going to be open sourcing models.

当然，很多产品工作本身都会与我们做的事情具有特定和综合的性质，因此，并不是我们所做的一切都会公开。显然，许多工作需要以一种创造我们产品独特价值的方式来开发。但我认为，在基本模型方面，我期望我们会推动和帮助构建一个开放的生态系统，这是非常重要的事情。这听起来很像Meta会开源一些模型。

One more Zuckerberg quote. On the AI tools, we have a bunch of history here. So if you look at what we've done with PyTorch, for example, which has generally become the standard in the industry is a tool that a lot of folks who are building AI models and different things in that space use, it's generally been very valuable for us to provide that because now all of the best developers across the industry are using tools that we're also using internally.

再来一句扎克伯格的话。关于人工智能工具，我们在这方面有很多历史。例如，如果您看我们在PyTorch上所做的，PyTorch已经成为了业内的标准，许多正在构建AI模型和其他相关领域的人都在使用它，它对我们来说非常有价值，因为现在整个行业最好的开发人员也在使用我们内部使用的工具。

So the tool chain is the same. So when they create some innovation, we can easily integrate it into the things that we're doing. When we improve something, it improves other products too, because it's integrated with our technology stack, when there are opportunities to make integrations with products, it's much easier to make sure that developers and other folks are compatible with the things that we need in the way that our systems work. So there are a lot of advantages. But I'd view this more as a kind of back-end infrastructure advantage with potential integrations on the product side.

所以工具链是一样的。因此，当他们创造一些创新时，我们可以轻松地将其整合到我们正在做的事情中。当我们改进某些东西时，它也可以改善其他产品，因为它与我们的技术堆栈集成在一起，当有机会向产品进行集成时，更容易确保开发人员和其他人员与我们需要的东西兼容，以了解我们的系统工作方式。因此，这有很多优势。但我认为这更像是一种后端基础设施优势，具有产品方面的潜在集成。

But one that should hopefully enable us to stay at the leading edge and integrate more broadly with the community and also make the way we run all the infrastructure more efficient over time. There are a number of models. I just gave PyTorch as an example. Open compute is another model that has worked really well for us in this way, both to incorporate both innovation and scale efficiency into our own infrastructure.

但这个模型希望能够让我们保持在前沿，更广泛地与社区集成，并且使我们运营所有基础设施的方式随着时间变得更加高效。有许多模型可供选择，我刚刚举了PyTorch作为一个例子。开放计算是另一个模型，对我们来说也非常有效，既可以将创新和规模效率融入我们自己的基础设施中，又可以更好地与社区整合。

So I think there are incentives, I think, are basically aligned towards moving in this direction. Now, that said, there's a lot to figure out. So when you're asked if there are going to be other opportunities, I hope so. I can't speak to what all those things might be now.

我认为目前的激励措施基本上都是朝着这个方向前进的。但是，需要理清很多事情。如果你问我是否还会有其他机会，我希望是的。但是现在我无法说出这些机会会是什么。

This is all quite early in getting developed. The better we do with the foundational work, the more opportunities, I think, that will come and present themselves. So I think that that's all stuff that we need to figure out. But at least at the base level, I think, where we're generally incentivized to move in this direction. And we also need to figure out how to go in that direction over time.

这个事情正在开展的很早。我认为，我们做好了基础工作，就会有更多机会展现在我们面前。所以我们需要解决的问题很多。但是，至少在最基础的层面，我认为我们普遍都有动力朝这个方向前进。我们还需要逐步确定如何朝这个方向前进。

I mean, I mentioned Lama before. And I also want to be clear that while I'm talking about helping contribute to an open ecosystem, Lama is a model that we only really made available to researchers. And there's a lot of really good stuff that's happening there. But a lot of the work that we're doing, I think we would aspire to and hope to make even more open than that. So we'll need to figure out a way to do that.

我的意思是，我之前提到过Lama。同时，我也想要明确，虽然我在谈论帮助开放生态系统的贡献，Lama只是我们仅仅为研究人员开放的一个模型。在那里有很多真正有用的东西正在发生。但是我们正在做的很多工作，我认为我们会雄心勃勃地希望并且希望比那更加开放。所以我们需要想办法去做到这一点。

Forgive all the quotes. But I think the time that Zuckerberg spent on this point is potentially extremely important. First off, PyTorch is an excellent example of Metis successfully open sourcing infrastructure tools in a way that benefits the industry and, by extension, meta itself.

请原谅所有这些引用。但我认为扎克伯格花在这个问题上的时间潜在地非常重要。首先，PyTorch是Metis成功开源基础设施工具的一个很好的例子，这有利于整个行业，进而有利于Metis自身。

Second, note the phrase, quote, make the way we run all this infrastructure more efficient over time. End quote. As I noted, this was the biggest outcome of Lama's availability. And the biggest potential beneficiary of AI models running efficiently and potentially running locally would be meta. One of the points we made in the follow-up calls that all Metis cap-ex spending has been for AI training. Incorporating AI in a user or advertising facing way will also necessitate spending on inference capacity as well, which will be far more expensive given the size of Metis user base and the ongoing need for inference.

其次，注意这里的短语：“quote, make the way we run all this infrastructure more efficient over time. End quote.”正如我所说，这是拉马可用性的最大成果。AI 模型有效运行的最大潜在受益者将是 Meta。在后续电话交流中，我们指出 Metis 所有资本支出都用于 AI 培训。将 AI 应用于用户或广告展示方面还将需要增加推断能力的支出，鉴于 Metis 的用户基数以及不断增长的推断需求，这将是一个更加昂贵的开支。

That, by extension, means that no company would benefit more from increased efficiency than meta. And the ability to run models locally would be a total game changer, as Metis would get the inference capacity for free.

由此可见，没有任何公司比Meta更能从提高效率中受益。而能够在本地运行模型将会是一个彻底的游戏变革，因为Meta将获得免费的推理能力。

Third, in that last paragraph, Zuckerberg notes that Lama isn't truly open source, but they saw these gains all the same. And yet Metis quote would aspire to and hope to make its work even more open than that end quote.

第三，在最后一段中，扎克伯格指出拉马并不是真正的开源，但他们仍然看到了这些收益。然而，Metis的引用语希望比那更开放，并希望使其工作更加开放。

This makes me wonder if Zuckerberg sees Metis as one of the few companies that can actually make AI broadly accessible and that that is a good in its own right. I certainly agree and hope so.

这让我想知道扎克伯格是否把Metis视为少数几家实际能够普及AI技术、并本质上是一个好机构的公司之一。我肯定赞同并希望如此。

I am in the camp that believes the potential upside of AI is massive. And the way through the upcoming disruption is to accelerate and double down on innovation.

我属于认为人工智能潜力巨大的阵营。面对即将到来的变革，我们需要加速并倍加创新来应对。

Trying to put the cap back in the bag is impossible at this point, but I am still wary of a dystopian outcome where AI is only ever in the hands of a few companies and governments who will claim it is for our own good. Even as untold potential benefits are foregone, we are the most high tech version of central planning yet.

现在尝试把盖子放回袋子里是不可能的了，但我仍然担心一种反乌托邦的结果，即人工智能只能掌握在少数几个公司和政府手中，他们会声称这是为了我们的利益。即使我们放弃了无数潜在的好处，我们仍然是最高科技的中央计划版本。

The daily update is intended for a single recipient, but occasional forwarding is totally fine.

每日更新仅供单个收件人使用，但偶尔转发也完全可以。 意思是每日更新是为了一个人而制作的，但有时也可以转发给其他人。

If you would like to order multiple subscriptions for your team with your group discount, please contact me directly.

如果您想要使用团体折扣来为您的团队订购多个订阅，请直接联系我。

Thanks for being a subscriber and have a great day.

感谢您成为我们的订阅者，祝您拥有愉快的一天。