（预告）Sora 2 与 AI 视频热潮、Meta Vibes 发布一周后，以及 TikTok、太阳能和 iPhone Air 的问题
Sora 2 大概率达到了更站得住脚的“GPT-3.5 时刻”门槛：它不是一场 ChatGPT 级别的新事件，而是生成视频终于变成连贯、可消费的完整表达。 Ben Thompson 起初对 OpenAI 的类比嗤之以鼻，随后区分了 GPT-3.5 与 ChatGPT 本身：如今的短视频已经足够连贯，能让大众用户惊叹，尽管专家早已从 Google 的 Veo 系列中见过类似能力。“它已经完成到足以像 GPT-3.5 一样，成为有意义的东西。”
Andrew Sharp 认为它具备 ChatGPT 式的病毒传播力，却没有 ChatGPT 式的实用性。 Sora 2 的照片级画面、直觉式界面，以及科技早期用户之外的人群自发兴奋，都显得格外重要；但 Sharp 看了大约90秒就想离开，也找不到娱乐之外的日常使用场景。他的质疑是：ChatGPT 一出现就有用，而 Sora 目前看起来“毫无用处”——不过他明确表示，这种情况可能会改变。
OpenAI 可复制的优势，可能在产品化，而不只是模型质量。 ChatGPT、吉卜力工作室风格的图片生成时刻，以及如今的 Sora，已经构成3次病毒式发布；Thompson 对投资者真正有意义的结论是：“发生3次，就值得持续关注。”Sora 一度冲到 App Store 第3名，而 Meta AI 约为第125名；Gemini 虽然需要邀请，仍排在第1名。
AI 视频可能把创作者阶层扩大到足以带来数量级的内容冲击。 Thompson 提出的1%创作、9%分发、90%消费模型，可能会在视频生成不再是“一件让人头疼的大麻烦”后失效：创作者从1%增至2%，供给就会翻倍；如果扩大到3倍、甚至10×，此前被制作约束挡住的想法就可能浮现出来。结构性变化在于“把想法与想法的具象化拆开”。
Meta 的注意力版图，正面临一种无法简单复制到 Instagram 内部的潜在颠覆性形态。 Thompson 认为，Meta 的权力和广告业务都建立在注意力之上。TikTok 曾经夺走注意力，但 Meta 可以用 Reels 回应；如果 Sora 的势头延续，Meta 将面对两个更棘手的问题——Meta 的 AI 可能不够好，用户也可能不想在 Instagram 里被强行塞入 AI 视频。他的判断带有条件，但措辞很直接：“Meta 可能会遇到大麻烦。”
Meta Vibes 遭遇的敌意反应表明，对大多数用户而言，缺的不是视觉新奇，而是叙事。 听众批评 Vibes 既不好笑，也不性感，还缺乏故事；Sharp 补充说，它看起来也没有提供任何信息或指引。Thompson 把它比作屏保，反映的是他异于常人的品味；他承认，把无叙事的环境式影像当成普遍有吸引力，是“一个巨大的盲点”。
Sharp 把今天的 AI 视频应用视为尚未成形的原型，终点仍不清晰。 他把它们比作1993年或1994年的电脑：能让人窥见4年或5年后的可能形态，尽管他基本认同外界对 Vibes 的反弹。Thompson 重新审视了自己对 AI 生成世界的乐观判断，承认奇幻环境可能吸引一批坚定的细分用户，却未必能成为大众产品。
外界反复提出的批评是，前沿资源正被投入到加速注意力衰退的方向。 Andrew 读到了 Ashkay Kothari 的提问：为什么人们还要把“我们最聪明的人、数十亿美元，以及地球上最强大的 GPU”投入另一个为注意力衰退优化的应用；他指出，关于 Vibes 和 Sora，类似批评都在反复出现。
1. Sora 2 与“完整表达”的门槛
Thompson 起初对 OpenAI 将 Sora 2 类比 GPT-3.5 嗤之以鼻，认为这是一次大胆借用技术史上里程碑的尝试；随后他修正了自己：OpenAI 说的是 GPT-3.5 时刻，而不是 ChatGPT 时刻。这个区别让他的判断从怀疑转为有保留的认同。
他的历史框架是：GPT-1 和 GPT-2 展示了涌现能力，GPT-3 已经能够自洽，但随后会滑向“虚无”。GPT-3.5 仍然有缺陷，也会产生幻觉，但它能给出一段有起承转合、普通人可以消费的连贯回答。
Sora 2 如今在视频领域完成了类似的工作。它的短片已经形成“完整表达”，不同于最初的 Sora；Thompson 认为，后者更接近 GPT-1 或 GPT-2，尽管当时的演示极具冲击力，而且 ChatGPT 的创造者本身也带来了额外关注。
Sharp 的推荐信息流通过荒诞叙事捕捉到了这种跃迁：Ronald McDonald 驾驶汉堡车，在洛杉矶 110 号公路上高速追逐。内容本身很蠢，但画面的连贯性和低门槛让这次技术进步立刻变得可理解。
2. 大众的惊讶比专家的惊讶更重要
Thompson 自己诊断出的盲点，是长期暴露在技术前沿。由于 Veo、Veo 2 和 Veo 3 早已向他展示过高质量生成视频，Sora 2 并没有让他震惊；但那些下载了简单消费级应用的人，却反应道：“天哪（Holy cow）。”
他借 DeepSeek 来打比方。付费 OpenAI 用户早已体验过思考模型，但许多免费用户第一次通过 DeepSeek “卖萌式”的呈现方式看到了可见的推理过程，并意识到这不只是一个新奇的中国应用，而是一个明显更好的产品。
Sharp 作为播客里的“普通用户代表”，认为 Sora 非常重要：界面直观，输出大体上达到照片级真实，而且有几位不属于科技早期用户的人主动发短信给他谈论这款产品。过去2年的 AI 产品中，没有什么比它更接近 ChatGPT 式的消费级势头。
3. OpenAI 的连续产品发布可能扩大内容供给
Sharp 对 Sora 最积极的判断，来自产品执行力：Veo 和 Meta Vibes 都存在，但 Sora 让创作变得显而易见。Thompson 表示认同，并数出了 OpenAI 的3次病毒式时刻——ChatGPT、吉卜力工作室风格的图片生成时刻，以及 Sora——他说：“这不是只发生过一次的事情。”
两人的分歧集中在实用性上。Sharp 每天用 ChatGPT 处理家庭项目、回答关于幼儿的问题，但看 Sora 大约90秒就退出；一位朋友曾经使用到凌晨2:00或3:00，可 Sharp 仍然看不到从娱乐新奇感走向日常工具的清晰路径。
Thompson 的反驳值得保留：娱乐型产品经常被当成玩具，但“最有价值的公司”往往正是通过娱乐用户来获取价值。因此，Sora 缺乏生产力属性，反而可能增强、而不是削弱它的商业意义。
更重要的是，AI 可能改变1%创作、9%分发、90%消费的比例。如果更低的制作门槛把创作者比例从1%推升到2%，供给就会翻倍；扩大到3倍、甚至10×，则会把想法与具象化这些艰苦工作分开，释放此前受视频制作约束的创作者。
4. Sora 给 Meta 制造了棘手的注意力问题
Thompson 仍然坚持自己关于 Vibes 的一个判断：全 AI 信息流让观众可以放松，因为每条视频都明确是假的，不必为每段照片级视频是否真实支付认知成本。因此，他认为独立的 AI 信息流可能比把合成媒体混入现有信息流更容易消费。
Sharp 偏爱 Sora 那种照片级真实、立足现实世界的输出；Thompson 则觉得这不够有趣，更喜欢 Vibes 的未来主义环境。这种分歧进一步凸显了 Thompson 的不确定：AI 生成世界究竟能否吸引大众，还是只会服务于一个细分群体。
OpenAI 也通过单独推出 Sora，避免打扰 ChatGPT 低摩擦的对话界面。Meta 选择相反路径，把 AI 视频放进 Meta AI，作为推动安装量的战略尝试，本身有其合理性；但 Sora 仍然冲到 App Store 第3名，Meta AI 约为第125名。Gemini 依然排在第1名，不过需要邀请。
Meta 历来通过把 Reels 放进 Instagram 来回应 TikTok。如果 Sora 持续夺走注意力，Thompson 认为 Meta 会遇到两个结构性障碍：Meta 的 AI“就是不够好”，无法与之匹敌；用户也可能不想在 Instagram 里被强行塞入 AI 视频。“注意力是一种稀缺资源。”
5. Vibes 难以留住注意力，因为奇观没有叙事
听众强调，Reels、TikTok 和 YouTube Shorts 往往提供幽默、性感、信息或叙事弧线，而 Vibes 只是用弹跳音乐承载超现实视觉。有听众称，这又把人类推向了“原子化、孤立的体验”，目标是变现注意力。
Thompson 通过自己的出版经历接受了这一批评。他曾经关注过一位技术能力很强的分析师，但对方的内容过于枯燥，始终无法吸引订阅者；Stratechery 的成功，部分原因在于最好的文章会提供主角、开头、过程和结尾。“缺乏叙事或讲故事能力，对大多数人来说没有吸引力。”
他的品味确实异于常人：未来主义的环境景观和 Apple TV 的航拍屏保，能接入他脑中“24/7”持续运行的内部叙事；外部强加的故事反而可能让他感到疲惫。因此，Vibes 对他来说像是去亲近自然——“我喜欢人造草。能怎么办呢？”——但这种偏好也造成了“一个巨大的盲点”。
Sharp 承认自己没有及时反驳：Thompson 在演示时已经注意到“Andrew 无聊了”，但 Sharp 仍然加入了热情推荐的行列。如今他把 Vibes 和 Sora 看成1993至1994年的电脑：尚未成形的实验，当前的笨拙感或许会在未来4年或5年、当终点逐渐清晰后，变得令人怀念。
6. 更广泛的批评是注意力衰退
- 最后的外部批评来自 Ashkay Kothari：为什么要把“我们最聪明的人、数十亿美元，以及地球上最强大的 GPU”投入另一个“为注意力衰退优化”的应用？Sharp 说，过去7天里，Twitter 上几乎每10分钟就会出现这个问题的不同版本，矛头同时指向 Meta Vibes 和 Sora。节目将其呈现为反复出现的观众担忧，而不是两位主持人已经确认的结论。
Hello, and welcome to a free preview of Sharp Tech. Hello, and welcome back to another episode of Sharp Tech. I'm Andrew Sharp and on the other line, Ben Thompson. Ben, how are you doing?
Feeling chastised, Andrew, but we can get to the reasons why momentarily.
Oh, boy.
No, I’m feeling old. That’s how I’m feeling.
Is this related to our emergency episode dedicated to Meta’s Vibes last Friday?
It might be. I told you—that’s why I don’t do emergency episodes. No, I think it’s a very ripe topic for discussion, so I’m actually looking forward to this.
It sure is. It feels like we’re recording in the middle of the slop revolution here. Very exciting stuff. We’re going to get to the Vibes reactions later in the episode, but I’m going to start with the news of this week, which was the release from OpenAI of Sora 2.
1. Sora Reaches The GPT Moment
I’ll read from an OpenAI blog post titled “Sora 2 Is Here.” They write:
“The original Sora model from February 2024 was in many ways the GPT-1 moment for video, the first time video generation started to seem like it was working, and simple behaviors like object permanence emerged from scaling up pre-training compute. Since then, the Sora team has been focused on training models with more advanced world simulation capabilities. We believe such systems will be critical for training AI models that deeply understand the physical world.
“A major milestone for this is mastering pre-training and post-training on large-scale video data, which are in their infancy compared to language. With Sora 2, we are jumping straight to what we think may be the GPT-3.5 moment for video. Sora 2 can do things that are exceptionally difficult and, in some instances, outright impossible for prior video generation models: Olympic gymnastics routines, backflips on a paddleboard that accurately model the dynamics of buoyancy and rigidity, and triple axels while a cat holds on for dear life.”
I’ll stop there, but we can go back to the sentence, “With Sora 2, we are jumping straight to what we think may be the GPT-3.5 moment for video.” GPT-3.5, of course, spawned ChatGPT. Is this a ChatGPT moment for generative video in your eyes?
So I kind of poo-pooed this statement. It seems awfully audacious to claim that we’re going to recreate one of the seminal moments in all of technological history.
Yep.
And I think that’s probably still fair. I’m not sure if anything in our lifetime will ever quite match—
It’s a very, very high bar.
Right.
I think that’s worth noting, yes.
In fairness, though, to this post, it doesn’t say we’re jumping straight to the ChatGPT moment for video.
Mm-hmm.
Even though Sora has gone viral—it’s number 3 in the App Store. I think Meta AI is around 125 or something like that, which we’ll get to momentarily.
Poor old Vibes. Yep.
What they did say is the GPT-3.5 moment, and I think that’s a distinction that I failed to appreciate when I was writing about this.
Mm-hmm.
And it’s been a little painful. I almost did an emergency post yesterday because I was thinking, “I need to correct the record here.” But I’ll have to save it for next week and just marinate—
Well, that’s what the podcast is for.
Yeah. I know, but some people don’t do both. So what was the 3.5 moment? GPT-1 was amazing, and GPT-2—
Mm-hmm.
GPT-2 was, I think, the one that really caught my eye. Then GPT-3 was like, wow, it’s actually making sense for a while, and then it would trail off into nothingness after a couple of paragraphs. And the difference is no one was paying attention at that time, right?
Yeah.
People who were watching were very impressed, but it wasn’t a phenomenon. By the time Sora came out, it was, “Oh, Sora is coming from the company that made ChatGPT.”
Mm-hmm.
And so it was a huge deal. In that context, it was impressive, but there were a lot of problems and a lot of issues here. And frankly, as we’ve discussed, Google sort of blew them out of the water with Veo, Veo 2, and Veo 3, whatever it’s called.
So backing up, in that context, Sora was like GPT-1, or even GPT-2 if you want to be generous. It’s showing emergent capability that is going to be a big deal but is not really useful yet.
Yeah.
GPT-3.5 was useful. There were lots of issues with ChatGPT when it launched. It launched with GPT-3.5.
Mm-hmm.
But it was still incredibly impressive. It hallucinated way more than it does now. It wasn’t nearly comparable to GPT today, but it was good enough, and it was a finished thought. You got a cogent essay in response to a question, with a beginning, a middle, and an end that sort of all made sense.
Yeah.
And because of that, it was consumable and impressive to people in a way that GPT-2, even GPT-3, weren’t quite there.
Mm-hmm.
If that is the framing they were going for here—and again, I should give them more credit and read their words closely instead of thinking about ChatGPT, which I was when I critiqued this sentence—then, yeah, that’s a fair way to characterize what we got with Sora 2.
It’s a finished thought. You can create 8-second videos, or however long they are, that hold together and are impressive. One of my learnings from this week is something I need to be on guard about when looking at these products: because I’m immersed in this and looking at everything, I probably consistently underrate what’s going to be a big deal and what isn’t—
Mm-hmm.
Because I’m not surprised by Sora 2’s capabilities.
The capabilities, yeah. Okay.
What I’ve seen from other people using it is that it’s blowing their socks off. I’m like, well, this is—you know, I’ve been using Google Veo, for example. I’ve been making videos. I’m not surprised they can do this. There are some new things here that the other ones aren’t doing.
But people who are not tracking it closely download this app, and it’s like, “Holy cow.” This happened in January with DeepSeek, actually. There are lots of things that are interesting about the DeepSeek moment, but up to that date, OpenAI had thinking models where, instead of just spurting out an answer, the model stopped and thought about it and gave you a much better answer.
All the thinking models were only available to paying customers. So for a lot of free users, their first experience with a thinking model was actually DeepSeek. It wasn’t just the novelty of it being a Chinese app. They actually got a remarkably better experience when they downloaded this app, and it had a very cutesy way of telling you what it was thinking about. It would be like, “Oh, look at the model thinking,” or “The AI is trying to figure it out.”
Like, “Wow, AI can do this?”
Right. Right. And so I’m thinking about how to talk about this and write about this, and I really feel there’s an angle here, which is the new emerging blind spots for me with AI. One of them is this.
Mm-hmm.
It’s underestimating, just because you can fast-forward and see, “Okay, this is going to happen,” that most people aren’t thinking about that. And when it lands, don’t underestimate what a big deal it is to drop a finished product. Now, is Sora 2 finished?
Mm-hmm.
No. But it’s sufficiently finished to be something meaningful, like GPT-3.5 was. So actually, this sentence is very accurate.
Yeah. Yeah. Well, for people who haven’t seen any of the Sora videos, I wrote down a sample of my Sora feed as of Thursday afternoon. One video was Ronald McDonald in a hamburger car in a high-speed police chase on the 110 in Los Angeles.
Yeah.
Then there was Martin Luther King Jr. saying, “I have a dream that Xbox Game Pass was still $20.” There was a video of the movie Interstellar rendered with Minecraft animation. Another was Sam Altman acting in a movie where he says, “We were never OpenAI. Welcome to ClosedAI.”
And then there was Jesus Christ himself live-streaming, walking on water, and turning the ocean water to wine. Right before we came on to record, someone sent me a Sora clip of scenes from the movie Oppenheimer, except the characters are all SpongeBob characters. It feels pretty stupid to read all that aloud—
Well, it’s okay.
But the slop era is here, Ben.
So, no, we’re—
We’re all just—
We’re—
Sidling up to the trough.
We’re into learning number 2.
Mm-hmm.
I personally do not find the Sora feed very compelling. I find the Jesus ones particularly annoying. I get a ton of them for some reason.
Okay.
I don’t know. I don’t like the guy. I did like the Ronald McDonald police chase one. That one amused me.
It elicited a—
O.J. Simpson.
A half smile from me.
Yes.
As I was reviewing it before the episode here.
But we talk on this podcast all the time about how I am the weirdo when it comes to entertainment choices, right?
Mm-hmm.
I'm not the one who's watching a gazillion shows on TV, even the good ones.
Yep.
I will treat myself, after a long week, on a Friday night, to an hour of Reels time, but—
Uh-huh.
It is so addictive that I don't like to get in too deep. I try not to spend too much time there. I love Twitter.
Put the phone down for the next—
The—
12 to 18 hours—
The text sort of thing.
Go touch grass.
I do have a note on Twitter, by the way, and phone usage that we should—
Okay.
Maybe come back to in a little bit.
Eagerly awaiting that later in the episode.
Yeah. So whatever I'm interested in is probably not what's going to be broadly popular. Whatever I'm not interested in might actually be popular. I sort of know that. I've known that for years.
Yeah.
And it's maybe useful to be reminded of that sometimes. I still like Vibes better than Sora. I—
Mm-hmm.
Find it inspirational, interesting. I like these futuristic environments. I also liked Midjourney and the Discord. Now, the Discord thing was a big blocker. I never actually went there because it's such a pain to use Discord. But this ethereal, completely different world. I think someone called Vibes a screensaver.
Hmm.
You know what I love?
Screensavers.
The Apple TV aerial screensaver, where it's like—
I don't mind that either.
The world shots flying around.
Yeah.
I will watch it for a while. I find it very calming and a real wake-up call, honestly. This is an issue that I knew, but, yeah, I'm not—
Well, thanks to the millions of people who have been using these generative AI video products, Ben has realized or remembered that he's a weirdo in the midst of—
That's right.
The crowd there. Yeah, I think I have a few takes on what's happening and the ChatGPT parallels.
I need your takes.
You're the resident normie. I need to be educated by you, please.
Well, I have to guard against my own hater impulses on some of this stuff. But I think this probably is a technical breakthrough on par with what we saw with ChatGPT. I'm pretty blown away by the quality of these videos, and maybe that makes me the normie who would've been blown away by DeepSeek if I wasn't already subscribing to the advanced models.
Opening up the app, A, it's really easy to use and intuitive, and B, the quality is pretty high and pretty impressive. And it's photorealistic videos for the most part, which I gravitate toward as opposed to the Midjourney aesthetic and the kind of ethereal, dreamlike quality that you get with some of the other AI content, particularly Vibes.
And the virality I've experienced in my personal life over the last couple of days since Sora launched has included a few people who I would put in the normie category. They're not early-adopter tech obsessives texting me about how crazy this app is. That is much closer to ChatGPT than anything in AI that I've experienced over the last 2 years. So there's some real momentum.
Where I pump the brakes on the ChatGPT comparison, though, is the total absence of utility beyond entertainment. ChatGPT, I use every day for household stuff, asking about different projects I'm doing around the house, or asking questions about toddlers or whatnot.
The Sora app, I can't use it, and maybe this is just me. I find myself wanting to bail after looking at it for 90 seconds. I can't imagine getting lost in Sora for 30 minutes, 2 hours, whatever it may be. We have a friend who did that last night and was up until 2:00 or 3:00 AM on Sora, so clearly some people can do that.
But I'm struck by how useless it appears now. I'm not saying it's going to always be useless, but I do think that's an important distinction from what we saw with ChatGPT, where it wasn't just a parlor trick. People immediately found it easy to use and helpful in their daily life. I'm not sure where the path is for Sora to become that kind of tool for people in addition to entertainment. Does that make sense?
Well, I have 4 responses to what you dropped.
Okay.
Number 1, I do think the bit that I mentioned about Vibes, about it being an all-AI feed, is still right.
Mm-hmm.
There's a bit where you mentioned you like that it's photorealistic and real-world. You're also going in cognitively aware that this is all fake—
Yeah.
Which lets you just enjoy it instead of trying to parse, video by video, whether this is real or fake. So I think that part of the take is still correct—number 1.
Mm-hmm.
Number 2, I agree it's super easy to make. This is really impressive. It's so much easier than Veo 3, whatever it's called.
Yeah.
Making videos is super easy, and—
And by the way, that was what was cool about it for me. I was aware of Veo, Veo 3, and I was aware of Meta's Vibes, but this is just—oh, it's... If nothing else, it's a testament to what a great product company OpenAI is.
Yes. That is my exact take.
Yeah.
Not just a great product company, but they've done it again, to your point about being viral. They did it with ChatGPT, obviously. They had the Studio Ghibli moment with their image generation. They have this moment right here with Sora, clearly going viral. Three times is worth keeping track of. This isn't a—
Yeah.
This isn't a thing that happened once, and that is notable and important, and I think something that we'll circle back to.
The bit about being a good product company and being easy to make: I talked about the 99/1 bit, where 1% of people create, 9% distribute, and 90% consume.
Mm-hmm.
The pushback I would have on your question about productivity is twofold. Number 1, everyone always dismisses entertainment-only apps as toys, not interesting, blah blah blah blah blah. ChatGPT actually helps me get work done. Guess what the most valuable companies do? They entertain you.
Yes.
Help us shut off our brains.
That's right, and that is just a reality.
They monetize us one minute at a time.
Yeah.
And so I wouldn't dismiss it on those grounds. I would actually—
Mm-hmm.
That would excite me on those grounds.
Number 2, they may be undoing that ratio. AI can actually make making stuff so much easier that that 1%—
Yeah.
Expands. And if you think about it, even if that 1% just goes to 2%, that's a doubling. That's a doubling—
Mm-hmm.
Of the people making compelling content, and I think on Sora there's a good chance it goes higher than that. We give lip service to, “Oh, AI will make more people creators.” I'm not sure we've fully internalized the implications of that.
We're talking about at least a doubling, maybe a tripling. Actually, maybe a decilling. What is 10 times? What does that mean for the amount of content that's out there, the likelihood of there being some brilliant creator who was limited by the necessity? We were just futzing around with video before here, right?
Yeah.
Video is a big pain in the ass.
Mm-hmm.
And to actually— I wrote, when the image generation stuff came out, I talked about the separation, the breaking apart of the idea and the substantiation of the idea. And that's a really big deal, and it's one of those things where I look at myself, even though I wrote that 3 years ago, and I think that was a very prescient observation on my part—
Yeah, mm-hmm.
Did I fully internalize that? Maybe not, so that's a pretty big deal.
2. Meta Faces An Attention Crisis
And the implication of these 4 observations is that Meta might be in big trouble. The reason is, at the end of the day, all their power, everything they have, all their ad business—it's all predicated on attention. And if someone else is consistently taking attention—
Yeah.
When TikTok took a lot of attention, it hurt them badly. A lot of people on TikTok are not on Reels. But at the end of the day, they could wrap that up by shoving it into Instagram.
Yeah.
Can they? They have 2 problems if this continues to be a big deal. Number 1, their AI just isn't good enough. They can't shove it in. And number 2, does anyone want it shoved into Instagram? I—
Mm-hmm.
Like I said, I still think the take on AI—that a distinct AI feed is cognitively better to consume than a mixed one—remains the case. And it's a very disruptive bit where Sora gets to come in.
One of the questions was whether it was right to launch a new app or shove it into ChatGPT. I think the ChatGPT app right now on the phone is actually kind of annoying. Sometimes I want to continue a conversation, but you open the app and you're immediately in a new conversation.
Yeah.
But I think that's actually a really good trade-off they make, because the time to get into conversing with ChatGPT is extremely low. In this case, I think not messing with that, not trying to shove this in, is a good thing from OpenAI's perspective.
I understood Meta doing it the opposite way, because they just want people to install the Meta AI app, which no one has, and to potentially use it. But to pull this off and get the app to number 3 in the App Store—by the way, Gemini's still number 1, albeit it's gated by invites, so we'll see what happens there—Meta might be in trouble here. Attention is a precious thing, and the route to recapturing it in this space has structural factors working against them.
3. Vibes Faces Audience Backlash
Well, I got bad news for you, Ben. Meta's not the only one in trouble, because we have a sample of the Vibes reaction that we're going to read through now. We exhorted people to try Meta's Vibes on the Emergency podcast and said, “Email us. Email us your thoughts.” Let me tell you, Ben—
Oh, yeah, give it to me.
The audience responded to that prompt.
Yeah.
I'll read a sample. Okay.
Helps me stay humble. Give it to me.
That's right. It's good for Ben, absolutely. This is character-building stuff.
Lulian says, “Ben, I was very surprised by your Meta Vibes take. People are addicted to Reels because they are either funny or sexy. AI slop can be neither. Artificial aesthetics without something visceral will quickly become boring.”
And Carlin says, “Ben and Andrew, I listen—”
Yep, I just want to acknowledge: yep, it's not funny or sexy. I like watching Apple TV screensavers. I'm going to take the L on that one. Continue.
There you go. Plus one to Lulian.
Carlin says, “Ben and Andrew, I listen to the Emergency podcast, downloaded the new version of Meta AI, opened the app, and was hit immediately by a viscerally negative reaction. Most short-form video on TikTok, Reels, or YouTube Shorts has some kind of narrative arc, but none of the videos I saw in Vibes had anything resembling a narrative—just cutesy, surreal, or eye-catching visuals paired with bouncy music.
“The bull case is that creators can now use AI generation tools to create compelling narratives, but watching the initially seeded content, I was surprised that Ben was so thrilled by it. It's just one more step in the depressing arc of engrossing technology that drives our society toward atomized, isolated experiences in service of commodifying and monetizing our attention. Ben, go touch some Wisconsin grass.”
And Jeremy echoed that sentiment.
No, no, wait. If Carlin's going to come that hard, I'm going to come back in return. So here's—
By all means. The floor is yours, sir.
4. Vibes Lacks The Narrative Hook
So overall, what Carlin is saying is right. It's funny, this is a lesson I learned with Stratechery over the years. When I started and it was successful, some other people would start sites, and I thought they wrote really interesting things. I was surprised at how relatively unsuccessful a few of them were.
There's one guy that sticks in my mind. I don't even know what happened to him. I'm not even sure he's on Twitter anymore, but he had really good takes that I consistently learned a lot from, and he tried the subscription model, and it went nowhere. I'm not even sure if he got 100 subscribers or something like that.
Okay.
I was trying to figure out why I was successful and he wasn't. I realized that what he was posting, I thought, was very substantive. The analysis was very good. I was consistently evolving my thinking in reaction to what he wrote, and it was incredibly dry and boring.
No narrative hook. It's all about telling a story.
Exactly.
Yep.
The best Stratechery articles all tell a story. The reason why the main articles that are on the front page of Stratechery are so painful to write is that it's so hard to figure out what the narrative is. I know what I want to say from an analytical perspective, but for those ones, I want to deliver a narrative. I want to tell a story.
It needs to have a beginning, there needs to be some sort of protagonist, there needs to be a middle, and there needs to be an end.
And this was—I think, not to pat myself on the back—I didn't start Stratechery meaning to write this way. It just came fairly naturally to me, and that is unquestionably a critical factor in Stratechery's success, particularly in the early years. It would draw people in, and they'd want to read it, and it would feel very satisfying to consume it.
Mm-hmm.
I think there might be something similar here. It goes like the TV thing. I have a story in my head about the way the world works. Part of the reason—
Yeah.
I find TV shows tiresome is I'm like, “I could go to Wikipedia and get the story here.” I'm just not caught up in someone else telling you the story. I love sports on TV because there's drama. There's uncertainty. I don't know what's going to happen.
Okay.
I'd rather watch a bad baseball game than watch an episode of Succession, just because the potential for drama and uncertainty is what I find compelling. That's not in my head. That's something that I don't already have a thought on—the way this should go or what it should be.
And so when I would read this guy's analysis, I thought it was really good. I was reading it very actively, plugging what he said into my view of the world, thinking, “Hmm, that's interesting. I never thought about it that way,” et cetera. The narrative was in my head the whole time, so I didn't need his narrative.
Now, I'm not saying—
Yeah.
This sounds hard to say without sounding like I'm just not someone who wants other people's narratives.
Uh-huh.
For Vibes, I see some of these futuristic landscapes or these things, and I find them inspirational to my internal narrative. I'm thinking about this and what works. There's even an aerial Apple screensaver flyover of LAX airport, and I'm looking at it and looking at the traffic flows and how horrific it is, and where the planes park.
I'm just—it's hard being me. Whatever. My mind's just going a zillion miles per minute all the time, and I find the little narrative cutesy stuff—occasionally they're funny. I mostly just find it annoying in the videos. I'm like, “I have my own story. I don't want your story.”
Uh-huh.
And this was such a massive blind spot. I was totally wrong. Again, I'm not saying I'm better or worse than other people. Whatever I'm talking about is certainly a function of why I've been writing Stratechery for 12.5 years now, writing multiple times a week.
Mm-hmm.
People ask me, “How can you generate so much content?” This is the answer. The answer is that I'm generating narratives in my head basically 24/7, and it's almost like a relief to be able to get a few out a week—
You want your passive—
Sort of multiple times—
—consumption to be narrative-free. That's what I'm taking away from this digression here.
That's right. Yeah, I always want an escape from narratives, right? Or I want drama and uncertainty because that's interesting. Again, I'm not saying this is a better way. It's actually a very annoying and painful way to live in many respects. I'm lucky I got a job—
Uh-huh.
—that basically rewards this sort of mindset, but it was a massive blind spot. I think what Carlin is saying is actually totally correct. The lack of narrative or storytelling is not compelling to most people. I'm the weirdo—
Yeah.
—here. And so I think this was, again, I think this was—
Or even information. I think short-form video—I mean, a lot of people feel like they're learning when they're watching short-form video.
Right. I love cooking videos and stuff like that. I don't know how many cooking videos I've watched and never—
Totally.
—implemented. I don't know if I've learned anything, but I do like those—
Vibes is—
I'll tell you that.
—completely devoid of any illusion of teaching you anything about anything, which is part of the problem and part of why people found it so viscerally upsetting toward the end of last week.
Again, I'm very hesitant to explain this. I want to explain it because I think I was wrong, and so I want to be honest. I always try to explain my thought process about how I got stuff wrong. I worry—
Mm-hmm.
—that the way I present this makes it sound like, “Oh, you know, I think about stories. I don't care.” But it's not like that at all. It's—
No.
I mean, there's no question about it.
I'm the one that's wrong here, so—and it's funny because, for me, when Carlin says I need to touch some Wisconsin grass, this experience is more akin to touching grass because—
Hm.
It's like stepping back from narrative for me—
A refresh—
—is the relaxing—
—for your brain.
—refreshing experience, right?
Yeah.
So, yeah. I like artificial grass. What can I say?
Well, the Emergency Pod will live on in infamy as part of Sharp Tech lore. I'll just read one note from Alexander. He says, “Ben and Andrew, am I taking crazy pills? As a faithful Sharp Tech and GOAT listener, I immediately downloaded the Meta AI app and started flipping through some AI reels. After about 30 seconds, I closed the app, and I don't plan on opening it again for another 6 months.
“I can see Ben's vision of AI-created worlds that you can explore in VR, but an app which just serves me AI video slop set to catchy music doesn't do it for me. I don't want to watch an AI-generated slam dunk. I want to see a real NBA player do it. To Ben, do you think this product will be successful? To Andrew, I feel like Ben's excitement is what got you fired up to talk about the Meta AI app, but will you honestly ever open it up again?”
So, for my part—
Take some responsibility, Andrew.
I will say—here's the thing. I record a lot of podcasts, Ben. I offer up lots of different opinions on lots of different things, and the only takes I ever regret on a podcast are takes where I feel like I haven't—
You agree with me?
—accurately conveyed how I feel.
On the Vibes pod, you were in such a euphoric place. I didn't want to bring the party down by disclosing to the audience that, on the Zoom call where you were showing Vibes to me and Domin, you did have to stop at a certain point because you said, “Okay, Andrew's bored. Let's stop.” And that's how I feel. That more than anything captures my reaction to what's happening on Meta's Vibes app.
I think, in general, the way I've reframed it—because a lot of people were really, really upset about the Meta product, and I think that reaction is interesting in its own right—but for me, the way I've reframed this in my mind is I look at all of these video apps as if we're looking at computers in 1993 or 1994. Rather than looking at those apps and saying, “Is this useful? Does it matter? Is it dangerous?” I just think it's kind of an interesting glimpse at where things are right now.
All of the apps are horribly half-baked and full of unanswered questions about where it's going to be 4 or 5 years down the line. But I do think we'll be able to look back at this era as kind of charming, as every company is sort of clueless about how to do this. And so I'm enjoying that look into the process. I can't say that I was enjoying Meta's Vibes app all that much, and even 20 minutes after we finished recording, I was like, “What did I just endorse?” And so I do agree with most of the haters that emailed in afterward.
Yeah, real betrayal of your principles. I'm disappointed in you, along with Alexander. I do think, just to add on to this point, I've generally been optimistic about VR and this idea of AI-generated worlds.
Everyone should keep that in mind when I express that optimism. You mentioned that you liked the more photorealistic aspect of the Sora app, which I—
Yeah.
—personally found less interesting. And that speaks to the fact that maybe these AI-generated worlds are not, to most people, going to be interesting. To some people—again, there's this whole universe of people that sit on Discord and just generate photos and go through them, for Midjourney, I mean, and find it really interesting.
Mm-hmm.
So, it's not that there won't be a market here. There are people that like to read sci-fi, for example, and like to imagine these fantastical worlds and these different technologies. And guess what? I liked reading sci-fi, right? Like—
I was just going to say, I'm not really a sci-fi guy. And maybe that's—
No.
—the problem with me and Vibes.
No, and most people aren't. I'm not sure if that's a perfect analogy, but I think there is something to it. Maybe this speaks to some of the critiques about worlds in your heads, or whatever it might be. I think I'm a pretty effective grass toucher—
Mm-hmm.
—I'm not too concerned about getting lost in this stuff. But I don't know. Maybe I need to revisit a lot of my questions and assumptions. I just think it's really interesting.
I always say that the most fascinating times for me on Stratechery are when I get something wrong. And—
Mm-hmm.
—this has been a big one for lots of reasons, and it is making me think about lots of things.
Yeah. Well, as we continue thinking about where any of this is going—
Including your willingness to push back. You need to do better.
Well, I know. At the very end of the podcast, where you were like, “You liked it too, right?” And then we both endorsed it on the way out, and I was like, “Ah, I should've taken that opportunity to say I got bored on the Zoom call.” But such is life, you know? All part of the fun with the Emergency Pod.
Yeah, whatever. Okay, you know.
It's terrible.
Next one.
5. AI Optimizes Attention Decay
I will read part of a tweet here from Ashkay Kothari on Twitter. He asks, “Why do we keep dedicating our brightest minds, billions of dollars, and the most powerful GPUs on Earth to building yet another app that optimizes for attention decay?”
I include that tweet not because it's especially unique, but because it's a question that has recurred every 10 minutes on Twitter for the last 7 days. It's a question people have about both Vibes and Sora. All right, and that is the end of the free preview. If you'd like to hear more from Ben and I, there are links to subscribe in the show notes, or you can also go to sharptech.fm. Either option will get you access to a personalized feed that has all the shows we do every week, plus lots more great content from Stratechery and the Stratechery Plus bundle. Check it out, and if you've got feedback, please email us at email@sharptech.fm.