即时反应:全新的 Meta AI App 已上线,而且酷得惊人|Ben Thompson 主持《Sharp Tech》
- Ben Thompson 在看到重新上线的 Meta AI App 后,收回了此前对 Meta“停滞不前”的 AI 业务的批评,并认定它已经交付了“一款好产品”。 其中带着“炸裂音乐”的纯 AI 视频信息流,比 TikTok 或 Reels 更让他难以退出,也比他在 YouTube 上见过的任何 AI 视频产品都更有吸引力——不过,首发内容很可能是 Meta 预先筛选的。
- 这款产品最关键的选择,是明确隔离:信息流全部由 AI 生成,因此用户清楚自己进入的是什么。 Ben 表示,把 AI 视频与真人拍摄的视频分开,消除了一个他“此前没有意识到”的“认知负担”;Andrew 也同意,混合信息流会迫使用户不断确认每段视频是否真实。
- Ben 最大的判断是,Meta AI 是“第一个 VR 产品”,尽管它通过 App 交付。 如今 5–10 秒的奇幻视频符合现有算力条件;同一套产品逻辑将指向头显和可探索的合成世界,AI 驱动 VR 的路径因此首次变得清晰可见。
- 强烈的 Midjourney 式美学反映的是首发素材的选择,而不是未来规模化的“用户提示内容”(UPC)。 Ben 预计,听众听到这期节目时,信息流中大部分内容都将来自 UPC。即便如此,他仍认为产品已经具备鲜明的观点,而且“这已经是它最差的状态”,因为体验只会持续改善。
- Ben 表示,Meta 更希望先把 Meta AI 做成独立 App;如果做不到,信息流也可能变成 Instagram 的一个标签页。 Andrew 自首次发布、约 9 个月前起,几乎就没再考虑过 Meta AI;与之相对的是,Meta 愿意“直接全力推进”。
- 对投资者而言,逻辑是:新增信息流 → 新广告库存 → 更低的广告价格,以及更长的增长跑道。 Ben 将其与 Stories 和 Reels 联系起来:当时投资者在增长到来前“撤退”了;变现可能要等上几年,但 AI 生成视频中的可点击商品提供了清晰的终点。
1. Meta 上线新产品,Ben 被迫收回前言
- 这期紧急节目之所以出现,是因为节目组 90 分钟前录制的内容还在说 Meta 的 AI 业务没有方向、进展迟缓。Ben 的原则是:“如果你不得不收回前言,那就越早越好。”
- 重新上线的 App 直接进入类似 Reels/TikTok 的 AI 生成视频与音乐信息流。Ben 觉得它“很抓人”,认定这是“一款好产品”,并表示自己更喜欢它,而不是在 YouTube 上见过的 AI 视频体验。
- Andrew 最初是通过 Ben 的 Zoom 屏幕看到这款产品的,因为更新还没有推送到他的手机。关键限制在于,Meta 很可能筛选过首发信息流;Ben 预计,随着用户提示内容逐步增加,听众听到这期节目时,信息流中大部分内容都已经来自用户提示。
2. 全 AI 信息流消除了隐形的真实性成本
- Ben 的核心用户体验判断是,这款 App “彻底拥抱 AI”。知道所有内容都是 AI 生成的,用户就不用担心被欺骗,也摆脱了一个他“此前没有意识到”的“认知负担”。
- 他的类比是:Twitter 属于前倾式体验,而 Instagram 属于适合广告的后仰式体验。真人与合成视频混在一起,会迫使用户反复前倾、判断视频是否真实;Meta AI 则让他们“知道一切都是假的,然后放松下来”。Andrew 补充说,这种分离也可能让用户更愿意观看以真人内容为主的 Reels。
3. 短时长合成视频照亮 Meta 通往 VR 的路径
- Ben 最激进的结论是明确的:“这是有史以来第一次真正的 VR”(This is real VR for the first time ever)。信息流让用户沉浸在完全虚构的场景中,却暂时不需要佩戴头显。
- 这些未来感十足、风格高度鲜明的视频带有“强烈的观点”,呈现出 Ben 所说的 Midjourney 式美学,也“到处都有 Daniel Gross 的手笔”。现有算力目前可以支持 5–10 秒的视频;从这里出发,Ben 认为下一步可以快进到由头显生成可探索的奇幻世界。
4. 先建立独立 App,再考虑分发到 Instagram
- Andrew 说,自首次发布、约 9 个月前起,他几乎没再考虑过 Meta AI,并问这套信息流最终是否会归入 Instagram。Ben 猜测,Meta 更希望先把独立 App 做起来;如果做不到,它也可能成为 Instagram 的一个标签页。Meta 会做“任何必要的事情”来确保成功。
- Ben 对算力成本持保留态度:一段视频可以只生成一次,之后反复用于信息流,因此逐步扩容可能有帮助。但他仍把这次由 Meta 筛选内容的首发视为底部,称“这已经是它最差的状态”,因为用户提示内容会不断增加。
5. 独立信息流创造投资者此前缺失的广告库存
- Ben 的投资逻辑始于 Meta 对新增广告库存的需求。库存越多,单条广告价格越低,广告就越容易被更多买家负担,同时也为未来增长留下更长的跑道。
- 他将这一机会与 Stories 和 Reels 相提并论:两者都增加了广告库存,投资者没能理解这一转变并“撤退”,随后增长大幅加速。与真人视频分开的 AI 信息流,可能同时改善两种体验,并让变现效率“显著提升”。
- 在 Meta 的推进顺序中,变现排在最后:先把 App 做出来,找到产品市场匹配,做大规模,然后再变现——可能要等上几年。AI 生成视频中的可点击商品提供了最终的广告路径;Ben 则提出用“UPC”,即用户提示内容,来定义这一新的供给类别。
Hello and welcome to an emergency episode of Sharp Tech complete with fake tornado sirens. I'm Andrew Sharp.
I'm honored, is what I am. I did a fake siren from a goat; it has made it to Sharp Tech.
Oh.
That's right. That's how you know it's an emergency.
Ben, we finished recording about an hour and a half ago. You ordered me back into the studio so that we could reconvene. Tell me why we're here. What's the emergency as we record today?
Well, number one, before I get to that, Meta, stop watching stuff on Thursdays. You do it all the time. It drives me bonkers. We could—
In a row. Two weeks in a row, throwing a wrench into our schedule.
We could wait till next week, except that what I want to talk about is directly applicable to the podcast we just recorded, and I might have to eat some of my words. If you have to eat your words, do it sooner rather than later. That's some life advice right there.
Maxim for life. Yes. For anybody who hasn't listened to the full episode Ben and I just recorded, there was a period in there where we talked about Meta's languishing AI efforts and the sort of rudderless ship over there with respect to AI, particularly the delay in shipping. But it does appear that they just shipped a new AI product that is directly relevant to what we were discussing.
So, at the same time that I had that weird interview with Mark Zuckerberg, they released the Meta AI app. I used it for about 90 seconds. The big thing was that they went to voice-first instead of chat or whatever.
I downloaded it. I can't say I used it very much.
1. Meta Launches A Compelling AI Feed
The Meta AI app has been relaunched. You open it up and you are immediately in a Reels/TikTok-style video feed. That video feed—turn on the sound. If you don't have it, update your app or delete it and reinstall it. I think that will get it to work. It is all AI-generated video with banger tunes.
I think what's in there now is probably all Meta put in there. If there is a user-generated content component, I'm speaking as if it's literally what Meta watched and put out there.
Yeah.
It's awesome. It's so good. I had a hard time tearing myself away from it, in a way that Reels and TikTok are insanely addictive.
Mhm.
They're insanely addictive in a lot of the Twitter sense that we were talking about earlier. There's just stuff in there that's compelling, and you're flipping through a bunch of crap.
Yeah. It's passively addictive, not necessarily engrossing, but you just find yourself scrolling and scrolling.
This was—again, it's launching with what was there to start. The UGC is going to be ramping up. By the time you hear this, it's going to be mostly UGC stuff. All AI video, and all of it was gripping.
Number one, this is a good product. I'm making the ruling right now. They shipped a good product, and it's a better and more compelling video AI product than anything I saw from YouTube.
Wow.
2. AI Content Removes Cognitive Load
Number two, I think right off the bat it seems obvious this is the right route, in that there's not real content in there. You know exactly what you're getting. It's not mixing stuff together. This is an AI product first and foremost. It's right up there, and I actually like that.
I'm not in there thinking, “Oh, is that an AI product or is that real?” You take away an uncertainty in my mind that I didn't realize existed when going through content, some AI and some real. I know I'm going in to get AI content, and that actually lets me enjoy it much more.
Distinction. Yeah.
Yeah, I'm not getting tricked or worried about that. That was a cognitive load I didn't realize I had until I experienced this app.
Mhm.
3. AI Video Becomes VR
Number three, and this is the big thing: my immediate feeling and response here is that this is the first VR product.
Wow. Okay.
It is virtual reality. It's all made up, and it's engrossing. You're in this little world and going different places. Again, the tuning is what they wanted up front. This has Daniel Gross's fingers all over it. If you go back and listen to some of our interviews, you know what he wants out of these sorts of products.
What do you mean by the tuning? What does that mean?
Just the sort of images that are in there. It's futuristic. It's very beautiful. It has a strong point of view. You could definitely see that in the Midjourney partnership that was announced a little bit ago. You get a Midjourney aesthetic. It's a product with a strong point of view.
We mentioned it in the podcast we just recorded. They kind of just stuck on Discord. They didn't want to be a big company. You feel like this is Meta's—or Meta and Daniel's; I'm going to give them the credit—saying, “Look, if you don't want to take this to the world, we will do it.”
The images I saw, or the videos I saw, are highly stylized. It does feel like AI, but it's very cool and bracing. The music does add to it.
That's right. It's leaning into being AI. The reason to rerecord here is, number one, I was insulting Meta a lot, and apparently they just turned around and dropped this amazing product. So, sorry, Meta.
Number two, it is a total affirmation, I think, of the whole thesis around AI and video. Again, we already talked about it in the last podcast. I'm the weirdo. I'm the Twitter obsessive. I like text, all that sort of thing. I found this so much more engrossing than I ever have TikTok or Reels. It was hard to tear myself away.
And again, this is the danger, which is a trap I've warned about. I usually get myself in trouble when I do emergency writing or respond right away. I should stop and think about it. So take all that I've said with that caveat in this emergency—
The emergency pod's fun, just for the record. This is a 24-hour reaction that we're hearing here. But I think the distinction between AI content merged with regular user-generated content—human content, let's call it—is important. When you say “user-generated content” in Meta's AI app—
4. User Prompted Content Emerges
You're right. There's user-generated content where they're filming themselves. It's real-world filming from a camera, versus what I'm talking about, which is users typing in prompts.
Yeah, exactly.
That's right: user-prompted content. Maybe it should be UPC.
Yep. Yeah, that is the distinction that I think is going to be resonant, because that makes more sense than asking people to go to one app and then sit there trying to figure out whether they're watching humans or AI the entire time.
And a great product is one where it comes out and it seems so obvious. In this case, the obviousness is, yes, I don't want it mixed. That was a cognitive load that was actually diminishing the quality of the experience with Reels or TikTok, because every few seconds or minutes you're like, “Wait, is that real or not?” Then it takes you out of the entire experience.
I compared it to Twitter in the past, where Twitter is a lean-forward experience. You're at work. You're out there. You're engaging. Instagram is a lean-back experience, and that's why it's great for advertising. You're just more ready to absorb new information. In retrospect, mixing these two meant you kept having to lean forward: “Wait, is that real or not?” or whatever.
This is a—I can lean back knowing it's all fake and just enjoy the amazingness of it. What's in there is so good. Again, it's the seeded content. So we'll see what happens with the—what do we call it?—user-prompted content, UPC.
Yeah.
But this is the worst it's going to be. By definition, all this is going to get better and better.
5. Meta Considers An Instagram Tab
Can I ask a question as a user? As a Meta user, I can't say I'm a Facebook user, but I'm a daily Instagram user. Is it possible that there's a future where this is a tab in Instagram, as opposed to me having to go to a completely separate app to access this TikTok-adjacent product?
Oh, definitely possible. Meta will do anything that's necessary to make sure that it succeeds. That's one of their strengths: they will just go for it.
Yeah.
But I mean, this bit about VR too. I mentioned it. I was so puzzled why Mark Zuckerberg didn't bring this up in our interview, and then—I think because I prompted him—he brought it up in the earnings call. He's talked about it again, about AI being the key to VR.
This is VR. This is what I'm talking about. You can fast-forward from this to having a headset on and being in fantastical worlds and exploring them, and it just happening. Right now, it's great for this product because the computer horsepower and the capabilities to do virtual worlds aren't there, but what is there is 5- to 10-second clips.
Yeah. It's pretty unbelievable. Was there any indication that this was coming? Because that's part of what's unbelievable to me.
No. Well, it was—
I haven't thought about the Meta AI app in months, probably 9 months, since they first announced it. That's why I ask about maybe putting this in Instagram. I don't think most people even know the app exists, so it's sort of an interesting curveball here.
Because I can't imagine the computational load this takes. Well, I don't know.
It’s interesting because it’s not generating it on demand all the time. People are generating it once, and then it’s in the feed, and they can use the content. So, regardless, I’m sure scaling slowly will be a good thing, not a bad thing.
I would guess Meta’s preference would be that this is so good, it actually establishes Meta AI as a standalone app. That’s possible. If that didn’t happen, then, yeah, this could be a tab in Instagram in the future. But I think the play right now is to establish the Meta AI app. Everyone listening should go download it just to experience it. It’s—
Exactly. Most people listening probably haven’t been in the Meta AI app for 9 months. Get in there and check it out.
It’s clear: if you open the app and it’s immediately in video, then it’s the right version, or you’ve been updated. If you don’t, you can try deleting and reinstalling it to see how it works. But I saw, like, a Meta shell here. But, like, this—it’s—it’s—
It’s really funny because you texted me and were like, “Download the Meta app.” I downloaded the Meta app—or redownloaded it. I deleted it and then downloaded it to try to get this update, and Ben has not updated on my phone yet.
So then Ben put it on the screen share of a Zoom call so that I could experience the new Meta AI app, because that’s how enthused you are by this development for Meta. We’ll see where it leads, I suppose.
Yeah, basically, immediately: one, this is real VR for the first time ever, and two, this cognitive-load bit about how this makes it so much more enjoyable. It feels like if you watch a product, you immediately just have total clarity on a couple of points. There’s something real there.
6. Meta Finds New Ad Inventory
Yes. Well, one other question: We talked a lot about the monetization for YouTube. This is essentially just an attention machine for Meta, is that right? And they splice in AI ads theoretically in the future?
Yeah. I mean, the Meta AI playbook—which I think I’ve talked about; I think I wrote this out ages ago—is now Zuckerberg’s talking point on every earnings call. First you build the app, you get product-market fit, you grow the user base, you get large, and then there’s a multistep process. Only at the end do you monetize it.
Mhm.
This is all definitely monetizable, right? You can have products that are AI-generated videos, and the products in there are clickable. We’re already in the world of AI-generated ads. So, yeah, there’s a path here, but that’s a path they, I’m sure, won’t tread for years.
Yeah. The only reason I asked is because we were talking about YouTube and Google, and you were juxtaposing where they are today with the Meta article you had written about 9 months ago, saying Meta has the opportunity to become the most valuable company on Earth if they take advantage of X, Y, and Z opportunities in AI.
The other thing—and this is why this is important—they have just launched.
It’s all part of the fun here. Yeah.
One of the big concerns about Meta—and why the AI bit about ads is so important—is that you need growing inventory. The more inventory you have, the more it decreases the price per ad, which makes it more accessible and gives you huge runway for growing in the future.
We’ve talked about this in the context of Stories, in the context of Reels. Both times, investors didn’t understand this; they bailed, and it actually led to massive amounts of growth going forward.
The reason why they will try to make this its own app is—well, actually, even if it’s a tab in Instagram, it’s a new feed. A new feed means new inventory. So that’s been the question: Where is their new inventory going to come from?
And, yes, if an AI video feed is totally separate from the regular feed, not only is that actually a better user experience, it’s dramatically better for monetization.
It makes both feeds much better. As for me as a user, I’m more likely to click into Reels knowing it’s all going to be humans. And then with the AI videos, we’ll see where it leads over the next couple of years.
But where they’re starting out is already very impressive based on what I saw on the Zoom call, where you insisted on sharing—
I am endorsing it. Amazing.
Yes, it is pretty interesting that it’s already—
It’s like Siskel and Ebert. If there’s a new product that both Ben and Andrew are enthused about, we’re going to find out. Does that mean—is that a bearish sign or a bullish sign?
An omen for Meta. Do you have any final thoughts before we sign off on what I think might be our first-ever emergency episode of Sharp Tech?
Well, I said we’re going to do some experiments with video, maybe being able to generate a new one at the drop of a hat. I don’t know. I was just going to say one more thing: I think UPC—user-prompted content—is going to be a term.
Newly minted on this emergency episode. All right. Well, on that note—
I’m pretty sure that alarm was actually set to end this emergency podcast.
Perfect. Great synergy.
Ben, stop shipping on Thursdays.
Yeah, I love it. I do. I love it. All right. Sounds good, Andrew. I’ll talk to you later.