Facebook is Dead; Long Live Meta, Does OpenAI Need to Log Off?, Questions on Bubbles, Blackberry, and Bell Labs
- Meta’s blowout quarter confirms the durability of its ad machine, but Ben Thompson thinks investors may now be crediting Zuckerberg’s superintelligence push for results it has not yet produced. Stories in 2017–18 and Reels in 2022 temporarily hurt monetization by moving attention onto under-monetized surfaces; both created inventory that Meta later converted into growth. This quarter’s gains came from recommendation systems such as Andromeda, addictive short-form video and ad-load optimization—not LLMs or superintelligence.
- “Facebook’s dead” because Meta is no longer meaningfully a social network; it is an entertainment company engineered to capture attention and serve ads. Only about 15% of viewed content now comes from friends and family, while TikTok taught Meta to recommend free user-generated content from across the network. That shift gives Meta more control than MySpace ever had: “A company that has dials is a powerful company.”
- AI is simultaneously Meta’s next product surface and a direct threat to the time its existing products monetize. Hours spent talking to GPT-4o companions are hours not spent in Meta feeds, yet generative AI also completes Meta’s long march from friends’ posts to globally sourced content and eventually content made uniquely for each person. Thompson calls ad delivery one of the world’s best existing agents: advertisers state the desired outcome, and Meta finds the customers.
- OpenAI’s GPT-5 launch became a test of whether the company can ignore vocal power users and operate a billion-user consumer product with conviction. The original automatic router elegantly exposed ordinary users to reasoning models without forcing them through GPT-4o, GPT-4.5, o3, o3 Pro, o4 Mini, GPT-4.1 and GPT-4.1 Mini. OpenAI then rapidly restored choices and complicated the picker, raising Thompson’s concern that it may be “a little bit too eager to please.”
- The defensible GPT-5 grievance was not personality or aesthetics but OpenAI taking paid capability away from Plus subscribers. Thompson estimates they went from roughly 2,900 potential thinking queries per week across o3, o4 and o4 Mini variants to 200 GPT-5 Thinking queries plus whatever the router granted them. His inference is that scarce GPUs were split between ChatGPT and a simultaneous API launch, “sacrificing these nerdy Plus users on the altar of an API” despite ChatGPT being OpenAI’s most defensible asset.
- GPT-5 Thinking is meaningful progress, not the AGI-like “Death Star” jump OpenAI’s own hype encouraged people to expect. Thompson says it is consistently better than o3 after roughly six months in his framing—more accurate, less prone to hallucination and better at following precise instructions—while maintaining his middle position that “people hyping AI are delusional, and people doubting AI are delusional.” Even with no further model gains, he believes the current product overhang could support a decade of development.
- An AI investment bubble is likely, but the enduring asset will be power capacity rather than fast-depreciating GPUs. Thompson distinguishes the dot-com companies from the telecom build-out whose bankruptcies left excess fiber for the next internet era; today’s productive analogue would be overbuilding electricity generation. With electricity cited as roughly 10–15% more expensive year over year and these companies willing to buy at almost any price, chip controls are “a total sideshow” beside whether the US can match China’s power supply.
- Across Apple, BlackBerry and Bell Labs, the recurring thesis is that control can create value—but concentrated institutions can also create assets markets would not fund on the same timeline. Cloud AI could spawn many cheap keyboards, glasses and earpieces without restoring BlackBerry-like margins; Apple may remain unusually valuable because of its 40-year product record, though Vision Pro arrived without a market “hole” to fill. Bell Labs likewise joined monopoly harms to extraordinary R&D, supporting Andrew Sharp’s preference for “purple” antitrust tension over absolutism.
1. Meta’s former crises were healthy monetization transitions
Thompson’s delayed earnings analysis became its own signal: earlier Meta disappointments demanded a “pants on fire, five alarm fire” response, whereas these spectacular results could wait two weeks. Even enormous AI spending now looks manageable beside the cash-producing core.
The repeated playbook is straightforward: “First we build a product, then we get customers, then we grow it, then we add monetization.” Stories in 2017–18 and Reels in 2022 pulled users from mature surfaces, causing impressions to explode while price per ad fell—ugly near-term accounting that established years of inventory growth.
Investors interpreted those periods as deterioration; Thompson saw eyeballs moving to less-monetized surfaces and thought, “This is incredible.” The ATT shock and publicly awkward metaverse pivot compounded the 2022 panic until Meta traded around $90, even though the underlying ad-network mechanism remained intact.
2. TikTok forced Facebook to stop being a social network
Sharp grants the old skepticism a logic: something that grows virally might also “die virally,” while Amazon can point to physical infrastructure. Meta answered that fragility not by preserving Facebook’s original form, but by repeatedly evolving Instagram through Stories, algorithmic ranking and Reels.
Thompson’s 2015 argument was that Facebook had to move beyond the social graph into personalized entertainment. He correctly saw the opportunity but wrongly centered professional creators; TikTok’s breakthrough was pulling user-generated content from anywhere, giving the platform effectively free supply and a stronger economic model.
The endpoint is jarring: roughly 15% of viewed content now comes from friends and family. Instagram and Facebook are therefore not social networks “in any meaningful sense”; they are entertainment feeds, a category Thompson considers more durable because content supply does not depend on one user’s friends continuing to post.
3. Meta’s quarter may reflect controllable dials, not superintelligence
Investor tolerance has inverted. Reality Labs lost another roughly $5 billion, putting it near a $20 billion annual loss rate—about double the $10 billion figure discussed during the metaverse panic—yet the market’s response to Zuckerberg’s spending is effectively, “Spend what you want. It’s all good.”
Thompson pairs incentive with what he calls a confession. Zuckerberg had an incentive to produce exceptional current results while announcing an all-in superintelligence push, while CFO Susan Li has explained that Meta learned to protect the short term so investors remain aboard for the long term. Otherwise, as with Intel technology promised for 2027 or 2028, investors can simply ask, “How about I buy your stock in 2027 or 2028?”
The quarter’s recommendation gains were real, but they came from systems such as Andromeda, a sophisticated model built on NVIDIA GPUs—not an LLM and not the superintelligence under discussion. Better relevance increased engagement, while price per ad held up better than the surge in impressions might ordinarily imply.
Thompson’s detective-story inference remains explicitly hedged: the cited shift toward 80% video, endlessly scrollable short-form content and “optimized ad load” may mean Meta increased both time spent and ads shown during the ideal quarter to do so. That is not proof of manipulation; it demonstrates that “a company that has dials is a powerful company.”
4. Facebook died when connection became incidental to attention
Meta’s 2016 posture was to cap ad load rather than diminish user experience, while Zuckerberg’s messaging increasingly emphasized serving individuals what they care about rather than strengthening community. Thompson’s conclusion is stark: “Facebook’s dead.”
Sharp hears the same change in Zuckerberg’s emphasis on serving individuals what they care about. Thompson goes further: the old language of connectedness may have obscured a product that was always “atomizing at its core,” with AI as the natural endpoint of individualized media.
That inversion changes the opportunity for shared publishers. Stratechery once differentiated itself as niche material against mass media; in a world of content uniquely generated for every person, Thompson thinks its value may instead be communal. Sharp’s joke lands the distinction: podcast listeners still hear the same mispronounced words.
5. AI both attacks Meta’s attention and perfects its feed
Thompson refuses to classify AI as only threat or opportunity: it is “a little bit of column A, a little bit of column B.” Time spent talking for hours to GPT-4o companions is time unavailable to Facebook, making ChatGPT more directly competitive with Meta’s attention machine than with Google’s ideal role as a launching pad to other sites.
The opportunity is the ultimate feed progression: first friends and family, then user-generated material from anywhere, eventually any content an individual might want generated on demand. Meta’s years of personalization make this less a pivot than the completion of an existing trajectory.
Zuckerberg’s strongest framing, in Thompson’s view, is that Meta already operates elite agents. An advertiser says, “I want more sales,” and the ad system autonomously finds customers; the news feed likewise searches for personally relevant content. Whether either uses an LLM is merely “a technical implementation detail.”
6. OpenAI needs consumer conviction more than Twitter approval
Consumer leadership is unforgiving at ChatGPT’s scale: a change that harms just 1% of one billion users produces 10 million angry people. Their visibility does not establish representativeness, yet their volume can make a successful decision feel catastrophic.
Thompson contrasts OpenAI with Facebook’s defining News Feed launch. Users organized online petitions and physical protests outside its Palo Alto office; Facebook issued conciliatory language but changed nothing because usage showed people loved the feed. Instagram later held its algorithmic timeline after seeing engagement rise by roughly 50%.
His own Stratechery lesson was that the same small group kept replying online while readership expanded rapidly, creating a structural mismatch between feedback and actual consumption. If OpenAI’s reversal followed usage data, Thompson accepts it; if it followed angry posts, he worries about its “constitutional capability to be an effective consumer app.”
7. GPT-5’s automatic router solved the mainstream-user problem
Sharp embodies OpenAI’s intended mass-market customer: he pays for Plus, opens ChatGPT, asks a question and accepts whatever model appears. The old picker—GPT-4o, GPT-4.5, o3, o3 Pro, o4 Mini, GPT-4.1 and GPT-4.1 Mini—was, to him, “Sanskrit.”
Ethan Mollick’s explanation supplies the product logic: GPT-5 is less one model than a switch selecting among models, sizes and reasoning budgets. Users stuck on default GPT-4o could finally see what a reasoner accomplishes without understanding why o3 might be more capable despite its lower-looking number.
Thompson, an almost exclusive o3 user, loved the original simplicity of GPT-5 and GPT-5 Thinking. OpenAI’s quick expansion to Auto, Fast, Thinking Mini, Thinking and Pro preserved routing but weakened the clean default—an aesthetically small reversal that Sharp considers strategically revealing.
8. Cutting Plus capacity gave every other complaint legitimacy
The substantive launch failure was a reduction in paid capability. Thompson estimates a diligent Plus user could previously distribute roughly 2,900 weekly thinking queries across o3, o4 and o4 Mini variants; GPT-5 initially left that user with 200 Thinking queries and whatever additional reasoning the automatic router happened to grant.
“That was crap behavior,” because OpenAI removed value from a $20 monthly subscription. Thompson calls taking capability from paying customers “anathema,” while noting that the most intensive Plus complainants may really have been Pro-level users trying to avoid the higher price.
His capacity explanation is pointed but inferential: OpenAI lacked enough GPUs while launching GPT-5 simultaneously in ChatGPT and the API. It therefore “sacrificed these nerdy Plus users on the altar of an API” exposed to intense model competition, weakening the consumer product and asset nobody else possesses.
Routing still points toward a powerful business model. OpenAI could spend more inference on a valuable legal query, produce a superior answer and attach lawyer referrals or affiliate links; free users would receive costly reasoning only when justified. The unresolved question is whether management can actually hold the trade-offs that model requires.
9. GPT-5 is roughly six months of progress, not a promised Death Star
Thompson’s durable middle position is that “people hyping AI are delusional, and people doubting AI are delusional.” Against claims of stagnation, he stresses that o3 arrived only about six months earlier and that GPT-5 Thinking now feels consistently, materially better—not an AGI discontinuity, but rapid improvement.
His home-improvement test was practical: o3 produced inconsistent guidance and unreliable links, sometimes ending at 404 pages. GPT-5 Thinking followed instructions more reliably, hallucinated less and assembled five relevant YouTube videos with the specific sections needed when no single video covered the whole job.
Sharp nonetheless assigns OpenAI responsibility for the disappointment. Altman posted a Death Star looming over Earth immediately before launch, after years of GPT-5 speculation; users expecting taxes completed, a polished app generated on the first attempt or novel scientific discoveries understandably felt they had been “promised flying cars” and received another app.
Thompson sees enormous product overhang regardless: even if models stopped improving today, current capabilities could set up ten years of technology development. A listener argued that OpenAI’s research decline began with Ilya’s departure; Thompson did not declare Meta’s talent war settled and emphasized ChatGPT, the consumer asset Zuckerberg most wants.
10. TikTok vibes may matter even when model-Twitter does not
Sharp’s prescription is blunt: “Double down, don’t apologize, ignore Twitter.” Thompson largely agrees, especially because model obsessives who cycle among Claude and every new release do not resemble people who may not even know Anthropic exists.
He gives OpenAI one caveat unavailable to Facebook’s earlier leaders. Modern short-form video can turn “GPT-5 is bad” into a mass-market meme; his son encountered that verdict without necessarily testing the product, and negative vibes can snowball into declining usage. “Twitter doesn’t matter,” Sharp concludes; Thompson answers, “TikTok might matter.”
The deciding evidence remains unavailable. A reversal could reflect normies leaving, power users shouting or executives fearing a self-fulfilling narrative; neither host claims to know. Sharp’s skepticism about data-driven reversals comes from OpenAI’s conspicuous attention to online conversation, including Altman’s stream of posts and the company’s repeated changes within days.
11. Model personality became product continuity for companions
One listener improved ChatGPT by storing a “helpfully contentious” instruction: stop praising him, identify flawed assumptions and correct misunderstandings. When shown dinner ingredients, ChatGPT responded that his supposed serrano was a poblano and would add almost no heat—exactly the useful contradiction he wanted.
Thompson uses custom instructions for terseness, prior knowledge before search and no reminders that the system is an AI. They help, but they do not erase a model’s deep tendencies; despite repeated demands, ChatGPT continued offering cloying compliments and affirmation.
Daniel Gross’s early analogy was that AI companies need rare “personality designers” just as computing needed exceptional interface designers. Claude 3.5 Sonnet had a distinct appeal absent from Anthropic’s larger model, suggesting personality can differ not only among labs but among models within one family.
For developers, relatively drop-in model replacement is good because surrounding scaffolding commoditizes the underlying supplier. ChatGPT faced the opposite problem: “the product is the model” for people using it as companion or therapist, so retiring GPT-4o meant “they changed their friend.” A future personality selector—and whether it can recreate traits such as GPT-4-powered Sydney’s antagonism—will test how controllable post-training really is.
12. The AI bubble is real, but its timing remains unknowable
Thompson thinks a bubble is “almost certainly” forming because transformative technologies requiring vast capital reliably produce one—railways, ships, electricity and the internet all did. But calling the dot-com bubble in 1996 would still have left years of upside: “You’re not right unless you get the timing right.”
Present demand is not imaginary. OpenAI has large, fast-growing revenue, and Anthropic has reached similarly large and rapidly growing revenue; Microsoft’s AI business and Google Cloud are expanding, while providers remain capacity-constrained. The uncertainty is how much demand represents durable deployment versus experiments or circular spending—Microsoft revenue downstream of ChatGPT, for example, or Anthropic revenue concentrated in Cursor.
Thompson’s empty Amazon envelope became the clean deployment specimen. A language-model chat accepted “I got nothing in it” and immediately shipped a replacement, restoring the frictionless service Amazon once delivered with humans before scale inserted cumbersome workflows. It was cheaper or at least better enough to create a happy customer who retold the experience publicly.
The dot-com era itself contained two bubbles: speculative internet companies and telecom operators laying broadband. Thompson says the telecom companies, including WorldCom, absorbed the deeper economic losses, but their uneconomic fiber became the durable foundation Google and the broader internet later exploited.
13. Excess power would make the AI bust productive
GPUs cannot play fiber’s enduring role because they depreciate too quickly. Thompson distinguishes roughly five-year GPU accounting—which may itself be too long—from data centers depreciated around 30 years; buildings and electrical infrastructure can host multiple chip generations, but compute hardware rapidly ages.
Sharp identifies the better analogue: power. If an overbuild drives developers bankrupt but leaves abundant electricity behind, the pain could resemble telecom’s collapse while furnishing the next era’s essential input. “This will be a productive bubble if we get power.”
Thompson cites gas prices falling about 10% while electricity rose roughly 10–15% year over year. Meta, OpenAI, Stargate and Microsoft will buy power at almost any price, letting utilities pass scarcity costs to everyone else unless generation expands. Against that constraint, NVIDIA export controls are “a total sideshow”: “China’s gonna have enough power. Are we?”
Deregulation and construction under the Trump administration were hopes, not accomplishments the hosts could yet identify. Thompson’s darkly optimistic target is “one of the most productive bubbles in history”: enough capital chasing generation that many builders fail, leaving society with cheap excess capacity.
14. Apple, BlackBerry and Bell Labs expose where durable value lives
Thompson separates advice to Apple from what benefits markets. Apple might rationally buy a model company, while the broader US could benefit from less dominance by a manufacturer deeply reliant on China. His default preference is often that incumbents return cash through buybacks so investors can fund startups better structured to innovate.
Apple complicates that rule through a 40-year record of making categories happen. Thompson’s answer to whether historical conditions or singular leaders matter is “yes”: Xerox PARC’s GUI, piracy and digital music prepared the ground, but Apple refined those trends. Vision Pro reverses the sequence—“All the Apple parts” are excellent, yet no compelling market hole surrounds them, and even Thompson had not worn his in weeks.
A cloud-centered AI world could produce an explosion of keyboards, glasses, earpieces and nostalgic BlackBerry-like clients because hardware becomes an accessory to the intelligence layer. That same modularity destroys BlackBerry-scale differentiation and margins; Thompson predicts Sharp would buy the keyboard device and abandon it “plus or minus three and a half days,” headphone jack notwithstanding.
Bell Labs captures the institutional tension. AT&T’s monopoly suppressed or delayed products, but funded world-class research; antitrust-driven patent access diffused Unix and other inventions, while IBM’s pre-emptive hardware/software split similarly mattered more than its eventual case. Sharp wants “purple” tension, not monopoly worship or absolutism—and Google’s slime-mold organization supplies the warning: autonomous fiefdoms can come together to solve hard problems, but move slowly and are “long-lasting and hard to kill.”
Verification Notes
- The transcript uses both “about six months” and, later, “six weeks” for the o3-to-GPT-5 interval; this digest follows the repeated roughly six-month framing.
Full transcript
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?
Doing well, Andrew. How are you? Back in the home studio, I see.
Indeed. Back in the home studio. A little disappointed, because last week we tossed around the idea of GPT-5 just replacing us, and we all just get to hang out and let the AGI take over.
Oh, that’s right. We are here. I guess that is the news itself: we are actually recording.
Back to the podcast mines, at least for another few weeks here. But it is good to see you, and we have a lot to get to. It’s been a busy week on Stratechery. We’re not even going to hit the NVIDIA news, but you wrote an article on Thursday, “Facebook Is Dead; Long Live Meta.” And so that’s where we’re going to begin.
1. Meta's Quiet Earnings Win
Yeah, this was a bit of an accidental article, actually. It was an update that I converted to an article. It’s been a very busy August, much busier than usual, to the extent that I’m 2 weeks behind in covering Meta’s earnings.
Mm-hmm.
And I was thinking about it when I was writing it. I thought, “Man, there are scenarios where it would have been a pants-on-fire, 5-alarm fire to make sure I wrote about these earnings, because the company’s blowing up.” In previous instances—for example, I remember 2017 and 2018, when Stories was out there and they had way more inventory, and their price per ad dropped. It was the first time their—or not the first time, one of the many times—their stock imploded.
I remember being in Wisconsin, outside on a beautiful summer day, and thinking, “This is ridiculous. I need to write about this.”
Drop everything. Yes.
Right. And so it was a sort of a meta-observation that it was okay for me to wait 2 weeks to write about these earnings, which were obviously spectacular, because they were fine. Even though they’re spending all this money on AI…
Mm-hmm.
It’s not a big deal. And that ballooned into, “Wait, there’s actually a larger observation in here.” So this was a bit of an accidental article, but I thought it was worth fleshing out a little bit larger than I originally planned when I started out.
Right.
And so that’s why we have an article on Thursday, the day of the podcast, with no reader emails. You’re going to have to come up with questions all on your own.
Exactly. Well, and it’s interesting, because I read the article, and initially I was like, “There’s nothing all that new here. Do we even have to add it to the podcast rundown?”
We don’t. We can skip it if you want. I know you want to talk about China.
No, but there is more going on than meets the eye at Meta. We’re recording this on Thursday afternoon. I’m going to read a piece of the article you published Thursday morning. You wrote:
“One of the weird things about Meta is how the company never seemed to get the benefit of the doubt, at least up until the last couple of years. There is something about social media that has always made investors intrinsically suspicious of the long term. Given that, Meta’s simultaneous slowdown in growth, mostly driven by ATT, combined with the public shift in focus to the metaverse, fueled concern that the company was desperately trying to pivot away from a failing business and losing billions of dollars to do so, with very few tangible results.”
Right. This is the other pants-on-fire episode, which was 2 or 3 years ago, when you had the same dynamic with Reels—
Mm-hmm.
—that was decreasing their monetization because it was a new product, taking attention away from their well-monetized products like Stories. And yes, everyone lost their mind.
Well, there were also the ATT headwinds, and the metaverse just looked pretty embarrassing in public. It all sort of snowballed into a panic. But you write:
“Things are very different now in 2 regards. First, there is the aforementioned commitment by management to deliver results now, not just promises about the future. Second, investor sentiment about Meta really has done a 180. If anything, my impression is that Meta is not just getting the benefit of the doubt, but actually getting more credit than they deserve.”
So tell me more about what’s going on and the credit that Meta does and doesn’t deserve in AI.
I mean, both of those panics—the one I call the Facebook Stories panic—
Mm-hmm.
—and basically the Reels panic—were driven, in my mind, by a real fundamental misunderstanding of Meta’s business, which I just referred to: this idea that it is actually healthy and bullish when Meta goes through these periods where monetization is hurting—
Mm-hmm.
—because customers have switched to the new product and monetization has not yet caught up. The way you see that is that there’s an explosion in ad impressions, because they have way more places to put ads, but the price per ad is just plummeting.
Yeah.
And investors get all worried about this. In both cases I’m like, “This is incredible,” because what is going to define the next X number of years? Them learning to monetize those spots and making all kinds of money.
Right.
This is a very organic growth story. It’s something that I feel like I laid out ages ago. I think this is one they adopted, and they’ve used it on their earnings calls all the time.
Mm-hmm.
We have a playbook: first we build a product, then we get customers, then we grow it, then we add monetization. I wrote that—the Facebook playbook—about a decade ago. I don’t know who was first, me or Mark, but they were doing it first. I don’t know if it was actually labeled as such, but that was the playbook.
And in both cases, in both scares, that was happening. That’s why the results looked bad. You had eyeballs moving from the monetizable surfaces to the less monetized surfaces, which hurt their income but was setting the stage for huge amounts of growth going forward. Investors were not giving them credit for that. That’s what I mean by not giving them the benefit of the doubt. They had done it before, they did it again, and investors should have been prepared for that. This whole scare was unwarranted.
I do think the metaverse/Stories scare, though, taps into something. I’ve remarked multiple times that Google’s the company I have a hard time understanding. Meta’s the company that makes me all the money.
Mm-hmm.
Because I feel like I get this company, and I’ve gotten it for a long time.
You and your readers.
Yeah.
A lot of people have done well on the back of—
Yeah, I mean, more than me, unfortunately.
Stratechery—
Yes.
bull calls, yes.
But this company’s been underrated from the beginning, and even in 2022, when this whole thing was going on, people were invoking MySpace.
Yeah.
No one knows what MySpace is, okay? And MySpace peaked at millions of users.
Well, that’s an interesting psychology lesson, because that line you had in there—“There’s something about social media that has always made investors intrinsically suspicious of the long term”—even makes sense to me. I wonder whether that’s just a relic of the early internet era and the way we came to understand this ecosystem, when you did have social networks come and go in cycles every few years. Even Facebook’s relevance waned and then was supplanted by Instagram, and then Instagram supplanted itself with Reels and Stories. It just continues to evolve. Facebook happens to be the one that’s driving all those changes.
But I think I understand why investors would at least be skittish, because with a social media network, it seems like there’s no long-term moat, whereas a company like Amazon, which has gotten the benefit of the doubt from investors for years and years, can just point to physical infrastructure as opposed to citing network effects or—
Yep.
—something like that.
2. Investors Distrust Social Media
Which I get. Yeah, I agree with you. There’s something about it that just makes sense: something can grow virally, and it can die virally.
Right.
It’s the assumption. It’s good you called out Instagram. I was going to link this in the article today. Maybe I should go back and find a place to stick it in.
Mm.
I wrote about how Instagram’s relentless evolution has dramatically changed what it was. I wrote this in 2021, when they were really leaning into Reels, and that was a big change.
Yeah.
But it was one of many big changes that Instagram has undergone from its beginning, and that’s been the app, to your point, that Facebook has used to stay on the cutting edge. I would quibble with your bit about Facebook pushing it. TikTok was a huge wake-up call. TikTok is where they nearly lost it.
And the reason they lost it—and I would say, of all the articles I've written, this is one of the ones I'm most proud of—is that I wrote an article in 2015, “Facebook in the Feed,” basically saying Facebook has to move past the social network.
Yeah.
There’s actually a larger opportunity to be your personal entertainment. I was focused on Facebook, which obviously ended up being Instagram more than Facebook, and I was focused on the opportunity for professional content creators, which was the mistake. What TikTok tapped into was just pulling content from anywhere across the network, so it’s user-generated content, which is obviously way better from an economic perspective.
Mm-hmm.
And it’s free.
Yeah.
But the overall point was that Facebook, in 2015, was running into disaster because it was too anchored on being a social network. It had to move past it. So maybe I was one of the investors not giving it the benefit of the doubt. But I feel like my angle was this: Facebook had this place in people’s lives that was exploitable and able to be more entrenched if it could break out of the social network mindset that it had.
Mm-hmm.
And that is what they’ve done. It’s pretty jarring. That filing I put at the end, with the FTC case, says that 15% of the content you see is from your friends and family now. It’s not a social network—
That’s crazy.
—not really in any meaningful sense. It’s an entertainment feed. It’s an entertainment company. And, by the way, entertainment companies have a pretty good track record of being pretty durable.
Mm-hmm. Yeah. I mean, they’re doing quite well. But I will say, when I began reading the article this week, I was bracing myself for an extended victory lap from you, where Meta is winning with AI today in a lot of the ways you predicted. But there wound up being a lot more nuance in both the numbers—
Andrew—
—and what you described—
Andrew—
—in the article.
As the chief zagger in the world, do you think I can appreciate your zagging if I don’t have an implicit streak of zagging myself?
Exactly. To that end, one thing that really entertained me was that the Stratechery tweet promoting this article was pretty cagey. You wrote, “Meta delivered blowout earnings the same quarter that Mark Zuckerberg doubled down on AI. I don’t think it was a coincidence.” And then, “An AI-generated bot. This has to be a bot,” was the first response to that article. “Meta’s AI push is definitely paying off. Reminds me of @GregoryBarnesx32’s take on how AI will reshape social media. Facebook might be fading, but Zuck’s bets are still hitting hard.” So—
Which had nothing to do—
That’s—
—with what I wrote.
That’s not quite—
That’s not even remotely related.
What’s happening there. I can understand why you would respond to that tweet and be like, “Yep, Zuck’s bets are paying off.” But again, it’s just a more interesting story as we sit here in 2025. So tell me about what Zuck is investing in and how correlated that is to the results we saw this past quarter.
3. Meta's AI Results Have Limits
The benefit of the doubt—this is where the benefit-of-the-doubt question comes in. Investors should have given Facebook more benefit of the doubt that it has this incredible ad business, that it’s going through what is actually a very healthy and productive cycle, and that it’s going to come roaring back in a big way.
Mm-hmm.
That was super obvious in 2017. It was super obvious in 2022, and yet somehow Facebook stock was $90, right?
Right.
So that’s where they didn’t give them the benefit of the doubt. This quarter, Mark Zuckerberg is spraying money all over the place—still not to Sharp Tech, for the record. But he’s spraying money all over the place. They’re spending all this money. They lost another $5 billion. They’re on a $20 billion loss run rate with Reality Labs.
Yeah.
Just an astronomical amount of money with precious little to show for it, to me, particularly given the length of time they’ve been working on it, despite the fact that the prototypes are super cool and all these sorts of things.
It is pretty remarkable. After the Metaverse freak-out a couple of years ago, it’s not like the losses have gone away in that area.
No, they’ve gone up.
Right.
They’ve increased significantly. At that time, I think they said, “We’re going to be losing $10 billion a year.” It’s now double.
Mm-hmm.
Inflation hits everything, I guess. But the results were amazing. They had their top-line beat, bottom-line beat, everything beat. It’s amazing, and investors don’t care. They’re like, “Spend what you want. It’s all good.”
Yeah.
Now they’re giving them the benefit of the doubt. And what’s interesting is, what it seems to me is, number one, you have an incentive. This is like a crime novel. You have an incentive to make sure you have good results in the same quarter you’re basically saying, “We’re going all in on superintelligence.” That would be a pretty good quarter to have super phenomenal results, right?
Yep.
Something they didn’t do before when it came to the Metaverse and things like that. Number two, you have the confession. You had Susan Li basically on with John Collison telling this story. She took over as CFO right around the time of the whole $90 stock thing.
Mm-hmm.
One wonders: maybe this is what drove the CFO change. Your stock gets knocked down to $90 because your messaging is wrong or whatever—that’s a reason for a change.
It’s a problem for anybody.
Yeah, that’s a reason for a change. An investor asked this question that hit me like a bag of bricks, which was, “If you’re spending on the future, how about we just invest in the future?” We talked about this in the context of Intel, right? The problem for Intel today is—or I think we said this a year or two ago, when things looked a little more optimistic—the problem is, oh, great, you have this technology that’s coming in 2027 or 2028. How about I buy your stock in 2027 or 2028?
Right.
Why do I want to be there in the nadir right now? And that’s a real problem for a public company, to have these investment cycles. She’s like, “We learned we need to take care of the short term as well as the long term to make sure our investors are along for the ride.” So we have the confession, and we have the incentive. And then what do we get last quarter? We get suddenly an explosion in ad impressions.
Mm-hmm.
And unlike 2017, when you had the introduction—or, like, the real spread—of Stories, which drove impressions organically, and unlike in 2022, when you had the real push in Reels that drove huge amounts of inventory organically for Reels, what happened last quarter? Why is there suddenly way more impressions?
Well—
Well, I—
AI is—
AI, the answer—
—is cited often. But—
So there is something to that, right? Their recommendations are clearly getting better, and that will drive price per ad.
Mm-hmm.
The price per ad did not decrease to the extent you would expect, given the amount of impression increase, which is in line with a few previous quarters. And they have these incredible new AI models that are not the AI we’re talking about.
Right.
Not LLMs, not superintelligence. They’re much more basic, simpler recommendation machine-learning models—much more advanced. Andromeda, which they talked about, is built on NVIDIA GPUs. It’s a pretty hardcore model, but it’s not the stuff we’re talking about.
Right.
That drove some of the results, for sure.
And it is stuff that you’ve highlighted in the past, like these sorts of capabilities should make their business more profitable in the long term as they lean into AI.
But even the stuff I’ve talked about with LLMs making AI better, none of that has come to bear yet.
Mm.
That’s still all in the future. All this stuff is pre-LLM capability coming to bear. So now, could AI make recommendations better so people use it more? For sure, yes. Even that is tied into this whole bit about suddenly it’s 80% video.
Right.
Short-form video is really freaking addicting, right? You sit down and you just scroll and scroll, and guess what? The better the videos, the more pertinent they are. Tighten up that signal, and you’re going to end up watching longer. Guess what happens—
Mm-hmm.
—when you watch longer?
More opportunities—
You see more ads—
—to push ads.
You see more ads.
Yeah.
And, by the way, they also mentioned that, “Oh, we optimized ad load,” a.k.a. I interpret that as, “We started pushing more ads to people.” The number of ads relative to whatever.
Right.
What it seems to me, again, if I’m a detective here, is that in a quarter where they really needed good results to justify this massive long-term investment, they might have turned a couple of dials to get some really good results.
And did it the old-fashioned way: pushing more engagement and pushing more ads, as opposed to leaning on superintelligence to supercharge the business.
The good thing about being an entertainment company is that you have an endless source of entertainment to push to people. If you're a social network, what are you going to do? Spur your friends to type more interesting things that you can push.
There's a bit where you can spin this super positively. This is actually a justification of why they deserve the benefit of the doubt. A company that has dials is a powerful company. The whole worry about a social network, and the whole MySpace concern, is that there's a certain bit that they don't have control over. It's like what happens—
If your friends leave—
If that happens to them.
If your friends stop—
Right.
Posting. Yeah, exactly. That's why those places didn't have a moat as opposed to Facebook. And so, in some ways, the results this quarter are downstream of Meta's recognition seven years ago that they needed to go in this direction.
Yeah, I mean, seven years ago might even be generous—maybe 4 to 5—but, yeah, this is my point. I'm like, "Where were all you investors all along, getting all worried at a time when they had just this organic, powerful network effect and ad network?"
And now today, again, I'm not denying the results. They're great results. I'm just saying it's pretty convenient they had good results, and it turns out I think they have more dials today to get good results than they used to. At the end, I'm like, "Well, maybe that's why." Maybe it's nice to invest in a company that has control of what it is.
It occurred to me—I just happened to link to that old 2016 earnings call. In that call—
Right.
They're like, "We're not going to increase ad load, so we're warning you impressions are going to stop growing because we don't want to diminish the user experience." Number one, that was an example of when Facebook probably did have a monopoly, because they could—
Indeed.
Facebook's dead.
Yeah. Well, I think at some point Mark talks about—look at me, a tech podcaster referring to Mark Zuckerberg by his first name—at some point Mark referred to serving individuals and what they care about, which is a pretty big deviation from a lot of the messaging for the first 15 years of Facebook's history, which always emphasized community.
Well, that's the thing with AI, right? AI is like the end game of all the internet, of this total atomization into the individual. And we talked about how this is why I'm optimistic about our business model, because we're a common piece of media when AI media is totally individualized.
Back in the day, Stratechery was very niche. It was, "Oh, I have content that's really relevant to me instead of the mass media that's everywhere." And a weird transition for me to think about with my business, and for us to consider as things we want to do in the long run, is that actually our opportunity is not to be niche. It's to be communal.
That's part and parcel of Meta doubling down on AI, and this is downstream of the feeds. The feed is the agent. That's one of the OG agents. Mark Zuckerberg made this point on a call, which is great. He got this one first. I wish I had. It was a great call: they have one of the best AI agents in the world, which is their ad-serving capability.
Yeah.
You type in, "This is the result I want," and Facebook's whole machinery goes off and gets you the result.
They go find the customer, yeah.
How is that not an agent? It's not using an LLM, but who cares? That's a technical implementation detail. It is a computer that you tell what you want, and it goes off and does it for you. It's like, "I want more sales," and it delivers you more sales. It's incredible. It's one of the best agents.
That's what the News Feed needs to make the same case. It goes out and finds you content that you're interested in, and it gives it to you. What has made Facebook unique, and the feed unique, is that what you see is different from what I see, and they do that at scale for every individual person.
There's a bit about Facebook—we're talking about the connectedness. Oh, this is how I should have framed it at the end: maybe Facebook talking about connectedness was this sort of subconscious pushback about how, actually, Facebook is atomizing.
Yeah.
It's not connecting. It's atomizing at its core, and there's a bit about this embrace of AI that is just an extension—it's a natural end state.
Well, first of all, I think we should record every podcast about an hour after you publish an article so we get to hear you second-guess different editorial choices you made live on the show. But, yeah, it's really true, and everybody who's listening to this podcast, I can assure you, you're all listening to the same podcast. You're hearing us mispronounce the same words. Ben, I believe, mangled Andromeda earlier in the show here.
Oh, is that how you say it? Yeah.
Yeah.
Pretty good guess.
I think so. I think it's downstream of a Stephen King novel.
Well, the funny thing is Passport is actually delivering everyone individual episodes.
That's true.
So—
But this podcast—
You don't know.
Identical—
We do, actually. We have the capability of delivering each and every one of you a different podcast, so you—
We can mispronounce a different word for every single listener out there.
That's right.
You'll never know.
As far as the transition is concerned, the other thing that was interesting to me reading this article is it seems like what's happening here is that there's now been an inversion of sorts. Instead of investors hand-wringing over the long term, it seems like Zuckerberg himself senses a long-term threat from AI, and so he's spending like crazy to mitigate that risk and, in the meantime, pushing more ads into Meta products in order to buy off the nervous Nellie investors. Is that a fair characterization of what's happening?
I don't know if it's fair. It might be right—
Okay.
—and also unfair. But, yeah, the AI-Meta thing is so interesting to think about. Is it a threat or is it an opportunity? It's definitely, I think, a little bit of column A, a little bit of column B.
Yeah.
The whole thing—we're going to get to GPT-5 and people with their AI friends, people spending hours talking to GPT-4o, which just sounds absolutely miserable to me.
Weird and very foreign, but apparently pretty common.
That's time not spent on Facebook. Remember, Facebook is a machine to capture as much of your time as necessary. In that respect, I think we mentioned this on a podcast a month ago—maybe actually Facebook is the one threatened by ChatGPT, not Google.
Google, over time, has become a property you spend more and more time on, and more of its money-making is on its site. But at its core and at its best, Google is a landing page, a launching pad. It's a place you go to go somewhere else.
Is that what AI is? AI is kind of self-contained, and in that respect, actually, it is a Facebook competitor, and so Facebook is threatened. At the same time, what Facebook has the potential to do—total individualization—is the end game of where Facebook and Meta have been going for years and years.
This is the ultimate feed, where the feed is not just limited to harvesting your friends and family. Now it's been expanded to find you user-generated content from anywhere. In the future, it can create any content you like. It's a natural sort of progression.
Yeah, I mean, there's definitely all kinds of opportunity for Facebook, and Zuck is going—
And also a threat. Yep.
Full steam ahead, so we'll see what happens on the superintelligence front. For now, though, speaking of superintelligence and AGI, we can turn our attention to GPT-5. I have a threshold question at the start here. There have now been several updates from OpenAI in an effort to satisfy angry users in the wake of the launch of GPT-5 a week ago.
Is OpenAI too online, and does OpenAI need to log off? The floor is yours. Your thoughts?
4. OpenAI Needs To Log Off
Who is OpenAI? I think that's the question. This is a real concern that I have for the company, to be totally honest. I think being a consumer product maker is really, really hard.
Mm-hmm.
It’s hard for all sorts of different reasons. One of them is that if you’re a billion-user company and you make a change that affects 1% of users negatively, that’s 10 million users.
Yeah.
That’s a lot of people who can be very vocal and can make themselves heard very, very loudly. I think there’s a certain approach to effective consumer product leaders, and here we go back to Mark Zuckerberg. The seminal moment—I always like to talk about the history of tech companies because there’s usually a moment early on that defines everything about the company going forward.
Mm-hmm.
For Facebook, that was the introduction of the News Feed. You start out with Facebook, and it’s just profile pages. You have to click around to all your friends to see what they’re up to. They come out with the News Feed, where all those updates are in one place and you can see it all, and people lost their freaking minds. There were online petitions. They had actual physical protesters in Palo Alto outside their offices.
My God.
They had to issue this mealy-mouthed statement about blah, blah, blah: “We respect…” They didn’t change a thing. They doubled down.
Mm-hmm.
Because what they did was look at the data, and they saw that people freaking loved it. They were on there all the time. There was actually a replay of this when Instagram shifted from the algorithmic timeline to the interest-based timeline, in 2016 or 2017, somewhere around there. Same thing.
There was going to be a replay of something like that with every major change that Facebook has made over the last 10 years, including Instagram in that umbrella.
Right. That was another one. By that point, I think they were more hardened. They were like, right away, “Actually, our user numbers show that engagement is up by, like, 50%, so—”
Right.
“We’re not changing anything.”
You guys are screaming into the void. Good luck.
Yeah. But it’s impossible to get good signal from the masses online.
Mm-hmm.
I’ve told this story. I had to learn this with Stratechery. Early on, especially when you start, you’re so eager. People are sharing my stuff on Twitter, and people are responding to you, and you’re replying. At some point, I realized that, number one, it’s always the same people replying.
Yeah.
And number two, my user numbers and readership are increasing rapidly. By definition, there’s a total mismatch between who I’m hearing from online, which is constrained and not really growing, and the people who are actually enjoying consuming my content. I need to be super careful about not over-indexing or paying too much attention to the loudest people on Twitter.
Totally. I’m the same way. I had the same experience writing, and I also sometimes have to step back and say, “Would I ever write angry feedback to a writer that I enjoy and read every day?”
Andrew, Andrew, Andrew. I am a listener of the GOAT podcast, where you recount getting J.J. Redick’s AOL screen name and messaging him every day to taunt him and troll him.
That’s true.
You messaged him every day to taunt him and troll him.
Wow. You just nailed me. So yes, when I was 16 years old, I was a big North Carolina basketball fan, and I did acquire J.J. Redick’s AOL Instant Messenger screen name. I harassed him on a daily basis because he had his away message up and was online all day. Eventually, I did have a conversation with J.J. where we traded insults in real time. It was great. But I’m saying, as an adult, I’m not going to be lurking in the comments section—
When did the adult line cross? I feel like you told another anecdote on GOAT, like, 2 episodes later, about just emailing and harassing someone online.
No.
You emailed—
No. Somebody else was trying to do that, and I advised that person not to email—
Oh, okay. Got it.
—to Jonathan Kuminga’s agent and tell him—
Okay.
—what a horrible job he’s doing, because he’s an adult and he’s not 16 years old. I said at the time, I would have done this when I was 16 years old.
Well, the problem is there’s that old New Yorker cartoon: “On the internet, everyone is a dog.” What you’re saying is—or, sorry, I got it totally wrong. Matt and I butchered that. The dog is sitting at the computer: “On the internet, you don’t know if you’re talking to a dog.” Whatever. You don’t know if you’re talking to a 16-year-old—
Or an AI bot.
—that’s the point you’re getting to.
Yeah.
Yes, exactly. Or—
Exactly.
Yes, my AI-bot repliers on Twitter today.
Well, and I’ll tell you why I ask the question: Is OpenAI too online? Does OpenAI need to log off? Of course, we could just be talking about Sam Altman being too online and Sam Altman needing to log off.
Oh, they’re all very online.
They’re very online. I had seen a former OpenAI employee talk a couple of months ago about how attuned they are to the conversation on Twitter and how conscious they are of the way they’re discussed on Twitter.
It was funny. Sam Altman and Elon Musk had a cat fight this week, or slap fight, or whatever it is. It was funny because I had to write an update about GPT-5—
Mm-hmm.
—and I had to link to, like, 18 Sam Altman tweets because he’s just on there all the time. So yes, I totally, totally know what you’re talking about.
Well, and you also had to do an update 12 hours later because they updated the model and the model menu for everybody. In terms of GPT-5 and what they announced last week, this is a topic where my sensibilities as a tech normie can actually be pretty useful, because I’m the target customer for OpenAI, not you. They want me and people like me to love their product.
As I was reading your GPT-5 article on Wednesday, I hit a point where I just threw up my hands, and it was like, “You’ve got to be freaking kidding me.” You had a screenshot of the legacy models list they have. Let me pull it up here. Hold on. It’s GPT-4o, GPT-4.5, o3, o3 Pro, o4 Mini, GPT-4.1, and GPT-4.1 Mini, and they have little explanations for what each model does.
You wrote, “As best I can tell, GPT-5 Pro is akin to o3 Pro, now with proper capitalization”—you put in parentheses—“and GPT-5 Thinking is akin to o3. I immediately defaulted to GPT-5 Thinking in line with the fact that o3 has been my go-to model for several months now.”
I’m sitting there as a normal person, Ben, reading that. All of it may as well be Sanskrit to me. It’s just completely out of control.
Oh, no. Are you outing yourself as a GPT-4o user?
I’m not a GPT-4o user. I use whatever GPT chooses for me. I open the app, put in my question, and get an answer back.
The answer’s yes. You’re outing yourself as a GPT-4o user.
Perhaps I am. I don’t know. But I’m a Pro subscriber—or a Plus subscriber.
Plus subscriber.
Excuse me.
Yes.
For the record, I offered you Pro. You turned it down, so.
It’s amazing to me that ChatGPT succeeded despite that confusion, and GPT-5 is an effort to make those choices much simpler for mainstream users, which is a great idea. To the extent they keep updating it and now further complicating the menu, it just seems crazy to me, and it seems like they’re listening way too much to a very small group of angry power users on Twitter.
5. OpenAI Sacrifices Its Core Product
Yeah. I’m actually personally very annoyed. I am a Pro user.
Mm-hmm.
I didn’t use o3 Pro, but I used o3 basically exclusively.
Yeah.
And so, no limits there was great. An amazing model. I actually think GPT-5 Thinking is a significant step up from o3 Pro or from o3.
Okay.
So I’m very pleased with this release. But now we have Auto, which decides how long to think; Fast, with instant answers; Thinking Mini, which thinks quickly; Thinking, which thinks longer for better answers; and Pro, with research-grade intelligence.
And not only that, but the original launch had GPT-5 and GPT-5 Thinking, all capitalized, by the way. GPT-5. It was so clean. It made me happy to go to the model picker and pick out my model. It was such a breath of fresh air after the ridiculousness before, and they’ve totally backtracked. I find it very upsetting from an aesthetic perspective.
But this is the big question I have: What Facebook learned is that you have to launch with a point of view and then—
Mm-hmm.
—you have to iterate. You have to have data feedback loops and base your decision-making on that.
Sometimes you do have to wind stuff back because the data shows, oh, we really screwed this one up. And by the way, that's something you might miss on social media, too. It may be the case that the normies are deserting, but all the nerds love it, and so you're hearing praise, but actually your product usage is falling through the floor.
Yeah.
Oh, there's a perfect example of this. I can't think what it is off the top of my head. Some startup, whatever, basically killed themselves chasing Twitter likes, and everyone just abandoned it. And so that's the key to the future—
You can also look at the last 15 years of media as an example of how that strategy—
Right.
—can go wrong.
Chasing Twitter likes, yes. And so did OpenAI. Was there a real shift in data—
Mm-hmm.
—and, like, “Uh-oh, we screwed up. We need to fix this”? I don't know. If there was, then all this backtracking and changes, re-adding GPT-4o, adding back all these choices... Now, there is one decision they made that was a very bad one, which we should get to in a moment. So let me—
Yeah.
—table that, to use some corporate speak. But if it was data-driven, then fine. I get it. Good job. You took a risk, people didn't like it, you backtracked. I get it. But if it was because a lot of people were mad online, that makes me very worried about your sort of constitutional capability to be an effective consumer app.
Yeah.
Which is to go out, have a point of view, make decisions, and be able to ignore the pullback because it's the right thing to do for the broader audience. And what GPT-5 was, I think, was right for the normie. It kills me that I'm sure I told you to use o3, and yet you are opening up the app and just using 4o. And 4o stinks. It stinks not just because it's an older model and super-optimized so they can run it cheaply and all those sorts of things. It doesn't think. Now, here we're using “think” in the AI term. Yes, it's anthropomorphized.
Yeah.
But we're gonna stick with it because I'm gonna keep saying it. This is the weird thing about the progression of models. It's like, oh, it's been 2 years since GPT-4. To me, it's been 6 months since o3, which is an incredible model. And the quality of output that it gets is so superior to what you were getting from 4o before then. No one was trying this. Number 1, free users didn't have access to it. Number 2, Plus users like you mostly didn't want to get hit by limits or whatever it might be. Who knew to go in there and pick out a model, especially—
Right, or you're just opening—
—because o3's a lower number than 4o. It's like, what's even going on here? So putting people in this, your default experience is going to, at times, use a thinking model when it thinks it's appropriate, and it's going to make that better. That is so clearly a better route for the normie, and it's still the default, but it's definitely softened, whatever it might be.
Yeah. Let me put a finer point on it for anybody who's not familiar with what was initially changed. Ethan Mollick writes, “A surprising number of people have never seen what AI can actually do because they're stuck on GPT-4o and don't know which of the confusingly named models are better. GPT-5 does away with this by selecting models for you automatically. GPT-5 is not one model as much as it is a switch that selects among multiple GPT-5 models of various sizes and abilities. When you ask GPT-5 for something, the AI decides which model to use and how much effort to put into ‘thinking.’ It just does it for you. For most people, this automation will be helpful, and the results might even be shocking because, having only used default older models, they will get to see what a reasoner can accomplish on hard problems.”
And so, as a user, I digest that with relief because I don't want the mental burden of having to know what each model does and means and having to determine whether I'm using the right model.
You don't want me in your ear lecturing you about how to use your computer. It's like—
Exactly. OpenAI, do it for me. You have my trust. And it seemed like that's what they recognized with the GPT-5 launch, and now it's all muddled and they've backtracked. I mean, you can still have it on auto for GPT-5, but—
It's not as clean, not as clearly the default. Now, where they did really screw up—
Mm-hmm.
—and this was a major screw-up that I think made all this so much worse, because there's a bit where they might have made some changes, and people always get upset at change.
Yeah.
But if it's intermingled with stuff you did that was actually wrong, it kind of gives validity to the whole thing, right? It's like, the reason why I was—not to get political—so anti-vaccine mandate for the COVID stuff was because I'm like, the problem is you're risking all the other vaccines—
Yeah.
—by conflating them all together, right? And so there's a bit... And why do I conspiracy-theory spiral? Because as soon as the myocarditis news came out—which, all in all, the numbers are still quite low—but suddenly it gives all these anti-vaccine cranks a point about which they were right, which then legitimizes all their other stuff about which they're not right.
Right. And I think the issue with COVID was, the vaccine wasn't 100% necessary for anybody who was under 60 or 65.
Well, they were mandating people who'd already had COVID to get the vaccine, which, if you know—
Right.
—2 things about what a vaccine is, it was nonsensical, right?
And so the skeptics—
There was so much that was done about it—
—can then say, “All these other vaccines also aren't necessary—”
Exactly.
—and they're pushing it on you.” It has in fact mushroomed and become a much bigger problem over the last several years.
Exactly. And so that was a very weird analogy for this bit about the release. If you were a Plus user like you—
Mm-hmm.
—who... Most of the Plus users complaining probably should have been Pro users. That's kind of the subtext of this, because they're like—
If you're nerdy enough to care.
Right. Because you had access to... You had, I think, 200 o3 queries a week. Then you had some number of o4 queries a day and a greater number of o4-mini queries a day. All these are thinking models. The point is, you had around 2,100, I think, or 2,900.
Mm-hmm.
I think it was 2,900. You had around 2,900 thinking queries a week. Now, you had to be a nerd and click your model, changing it all the time to stay within your limit. Like I said, you should've just got ChatGPT Pro, to be honest, or you should've gotten the Pro level. But when GPT-5 launched, suddenly Plus subscribers got 200 thinking queries and then prayed to the model router gods that GPT-5 would route you to a thinking model.
Mm.
That was crap behavior. You took away stuff from paying subscribers.
You're diminishing the value of my $20 a month.
Anathema. You cannot do that. You can't take stuff away. That was bad. It was a huge mistake. Why they made that mistake, I think, is very interesting, but that was in the midst of all these other complaints, and it gave validity to all the complaints. And so that's why, to me, that mistake is super interesting. Why do you take queries away from people? Well, it's expensive and you don't have enough GPUs. Guess who's been saying OpenAI is making a big mistake in their consumer business by not having enough GPUs?
You, for several years.
You wanted to be right, too, right? They launched GPT-5 on the API at the same time they launched it in ChatGPT. That meant they had to limit capacity somehow. They sacrificed these nerdy Plus users on the altar of an API that's going to be subject to intense competition, and they hurt their core, the real product, the real value, which is ChatGPT, to do it. Massive error. And the problem with this is: Is this a company that can make hard decisions? To be a good consumer tech company, you have to make hard decisions. ChatGPT was an accident. They're like, “Well, we watched it.” Usually, they're the accidental tech company.
Right.
Do they have it in them? Is Sam Altman the personification of GPT-4o?
Mm.
A little bit too eager to please, a little bit too eager to make people happy, not willing to draw the line and tell them the way it's going to be. These are the questions I have coming out of this. I think GPT-5 was in the right direction. It's clearly building to an advertising sort of model. When you start routing, you can route to all sorts of things. You can—
If the user makes a super-high-value query that you know they’re asking about legal stuff, for example, which might result in a referral to a lawyer, guess what? Spend the time to do a really good answer and, at the end, have affiliate links to lawyers or whatever it might be, right?
Mm-hmm.
They’re building the groundwork for this ad-supported model or affiliate-supported model that’s going to deliver. For free users, it’s going to be such a better product because they get the best model with the best thinking at the time that it’s appropriate, and they don’t have to pay for it. It’s great.
Yep.
But you’ve got to make trade-offs. Can they make any trade-offs at all?
Totally. Well, and I think you mentioned in your piece, Google and Apple give customers what they want, but they don’t ask customers what they want, and it seemed like there was some back-and-forth with the customer base that left the product in a more muddled place.
GPT-5 was pretty ballsy, what they did. Just: “This is the way it’s going to be. We’re dramatically simplifying. We’re going to make choices for you.”
It was ballsy but elegant—
But then they—
—and smart.
Exactly.
Like—
No, I think it was the right thing to do, and then they bail on it.
Less than a week. It’s been wild.
I had to wake up at 4:50 AM. I look at my phone: “Oh, they totally changed this. Cancel my distributions. Just churn out a whole extra update at the end of my update.”
Like I said, it’s been a busy week, man.
It has been a busy week.
Well, a couple of emails here. Anthony says, “Ben and Andrew, Ben pointed out a lot of reasons that folks were upset with the GPT-5 release, but I think he underestimated one important reason: expectations misalignment. A lot of people expected GPT-5 to be a massive improvement relative to o3. That expectation was grounded in hype around the release from OpenAI itself. GPT-5 is a big deal, mainly because it delivers reasoning to the vast majority of people who have never touched an o3-level model. It is not, however, the raw intelligence jump that many expected. As a daily o3 user, I found that GPT-5 Thinking is modestly better than o3 but certainly not a quantum leap. GPT-5 can’t do my taxes or vibe-code a perfectly polished app on the first try or come up with any novel scientific insights. My question is, do you think this release is indicative of model capabilities hitting a wall, or does this release show us that OpenAI is pivoting its resources away from striving for maximum intelligence and toward maximum ChatGPT usage?” What do you think, Ben?
6. GPT-5 Meets Reality
I think I feel really good about my continued middle-ground perspective, which is that people hyping AI are delusional, and people doubting AI are delusional.
Mm-hmm.
o3 was only 6 months ago. Having used it a few more days since I wrote about it, I think GPT-5 Thinking is actually a lot better than o3.
Okay.
Pretty consistently. In my experience, it hallucinates less, and you can give it really clear instructions. I was doing some home improvement thing, and I was using o3 a ton. It would generate weird links, and it would just not be consistent if I prompted it differently and things like that.
Mm-hmm.
GPT-5 Thinking is way more consistent and way more accurate. I can ask it, “Oh, get a reference.” It was talking about the wall-finishing sort of thing, and it was very hard for me to visualize. I said, “Give me some YouTube videos,” and it got me all these different YouTube videos and the different sections in them. It said, “I couldn’t find a YouTube video that perfectly captures what you want to do, but here are 5 YouTube videos that cover the different steps.”
That’s awesome if it’s giving you pin sites from the YouTube videos.
Right. This is actually where I found o3 falling down the most: in its linking out.
Hmm.
The links would work sometimes and sometimes they wouldn’t, and they’d just be 404s or whatever it might be. Again, I’ve only really used it for a couple of days, but in my experience, it is a lot better. And is that 6 months’ worth of improvement? Yes. That’s a massive improvement in 6 months, right? But at the same time, I totally get Anthony’s point. This is the too-online bit.
Yeah.
Just let it happen, right? Now—
Well, OpenAI also bears responsibility for the hype getting out of control. Sam Altman, the day before this release, tweets out a photo of the Death Star looming over Earth. We joke about it on the podcast, but that’s indicative of the way people have talked about GPT-5 for the past 2 years. So I understand why there are some people who are waiting to see AGI-like function and look at what GPT-5 is and are like, “Well, we were promised flying cars—”
Yep.
“And this is just—”
No, totally.
“…another app.”
It’s part of being too online. And—
Yeah.
By the way, I think one potential defense of OpenAI, particularly relative to Facebook—what did I say happened to Facebook when they did the News Feed? I said 2 things. This is a quiz. Were you paying attention?
Oh, you’re quizzing me? Protests outside the office and no changes on Facebook’s side.
And online petitions.
Oh. A relic.
Exactly.
Remember those?
So the other night, my son’s like, “So is ChatGPT-5 really bad?” And of course, Mark Zuckerberg is ruining my son’s life by giving him way too much short-form video, and I just have to pry it out of his hands. The vigor and intensity and tenor of online discourse is totally different from what Facebook was dealing with in 2009. If you have TikTok and it becomes a thing, it’s a meme: “ChatGPT-5 is bad.”
Yeah.
I can understand, number 1, why that can inspire a certain bit of panic, and number 2, why that panic might be justified, where you get a self-fulfilling prophecy spiral, sort of a snowball.
Vibes translate to tangible results in some ways.
That’s right. And so that does give me a little bit more understanding for all the flailing I’ve been doing for the last week.
I guess so.
That doesn’t give me any more understanding.
I think double down, don’t apologize, ignore Twitter.
That is true.
Twitter doesn’t matter.
That’s the real lesson.
Yeah.
But TikTok might matter.
Well, that’s true. But, yeah, I guess the conversation over the last week was all really, really crazy to me. Matthew asks in this same vein—
Oh, sorry. One more thing on there, too, by the way: the whole TikTok thing.
Okay.
Are the people consuming these videos going to switch to Claude? Do they even know what Claude is? This is the problem with being on Twitter. You’re with all the nerds who are testing every model the moment they come out.
Yeah.
They’re like, “Yeah, I mean, I haven’t touched Anthropic’s models for at least 3 weeks. I mean, they’re dead in the water.” These people are not—
This is what I think—
If you’re going to be a consumer tech company, yeah—
…ignore this community.
No. Mark Zuckerberg should pay me, and Sam Altman should pay you. I think that’s the takeaway we’re getting here.
Exactly. I’m happy to be a consultant. First lesson is going to be using capital letters at the beginning of sentences. Matthew says—
It looked so good: GPT-5 Thinking, GPT-5 Pro, all capitalized.
Finally—
It was great.
…capital letters.
Oh.
It’s great.
It looked so good.
Yeah.
Instead of—you go into that dropdown, you pick small o3. It’s terrible.
It drove me crazy reading your article this week. I was like, “I can’t believe people have been living like this. This has to stop.” Matthew says, “Ben and Andrew, can we already declare the Zuck talent war a victory? OpenAI’s research decline started earlier, mainly with Ilya’s departure, but just like Instagram Stories copying Snapchat, Zuck knows exactly when to step on the throat, and I think he’s succeeded in reducing OpenAI to probably 3rd place in the AGI/ASI race behind Google and xAI, in my opinion. What do you think?” Any thoughts?
Does he have Anthropic 4th or 5th? I mean—
I was wondering how Anthropic factors into the AGI/ASI race. Does that race really matter? And I’m certainly not declaring the Zuck talent war a victory because OpenAI—
No, that was my take.
is way ahead.
You totally stole it. I’m just not—I get that it’s a motivating factor. People think we’re going to get to this recursive improvement. Mark Zuckerberg talked about it on the earnings call two weeks ago.
I’m not there. I’m in the middle road. I think AI is a huge deal. I think the product overhang is huge. The things that can be built, I think, will drive the next—You could not have any improvements in AI today, and the next 10 years of tech development are set. All the stuff that is to be built.
And we are getting improvements, and I think the improvements are underrated just because we’ve become inured to what’s happening and how quickly it’s happening. There’s a bit where ChatGPT came out and then GPT-4 came out—they added it three months later. It was actually already built at the time.
That set these crazy expectations for this insane improvement. But I feel like the o3-to-GPT-5 jump is actually pretty similar. I think o3 is very underrated in terms of what a jump it was, and then this is only six weeks later—we’re in GPT-5. People’s sense of what’s possible is a little off.
But I am also not there that we’re still podcasting. I showed up on time. I was lying last week, Andrew, when I said we might not be here. That was all a put-on. I was all cap for me, as the kids say or whatever.
Yep.
Fronting for the listeners.
No cap from you. It’s not all cap.
Oh, no cap?
It’s misleading.
I can’t remember. I don’t know what it is.
But yes. You were 5 minutes early today. Early bird gets the worm for today’s session.
Yeah, I just don’t know where the AGI/ASI race conceivably is supposed to lead.
It’s very possible that OpenAI got hurt very badly. I also think that OpenAI’s biggest asset is ChatGPT, which no one else has, and Mark Zuckerberg would love to have.
And that’s why Zuck is spending.
Yeah, exactly.
That’s the gap he’s trying to close, and that’s not gotten any narrower in the last couple of weeks. Time will tell, I suppose.
Well, no, maybe it did get narrower. Maybe ChatGPT lost half its users overnight. We don’t know.
We haven’t seen the data. Yes.
That’s right.
We haven’t seen the data. It’s because they’ve been so publicly reliant on Twitter. That’s why I’m skeptical that there was data that swayed them toward those changes this week. Or not reliant, necessarily, but beholden to Twitter vibes. That’s why I’m skeptical that there was data driving them toward those changes. But who can say what was driving the decision-making at OpenAI?
One more ChatGPT note. Harrison says, “I gave ChatGPT the following prompt yesterday afternoon, and I have absolutely loved its results so far: ‘Add this to your memory,’” he told ChatGPT. “‘I do not want your responses to be overly complimentary. I’m smart, and I know I’m smart, so I do not need you to tell me when I’m being smart. Instead, I need you to tell me when I am misunderstanding something. Make sure your responses helpfully contentious. Do not be argumentative just for the sake of argumentation, but do point out when I display flawed assumptions.’
“Last night, while making dinner, I gave ChatGPT a picture of my ingredients and asked it how many serrano peppers I should use in the recipe. I’m a dreadful cook. Its response was, ‘That’s not a serrano pepper; it’s a poblano pepper. If you use it in place of a serrano, your dish will have almost no spiciness unless you add another heat source.’ What a helpful correction. I was so happy. It told me I was being an idiot chef because I know I’m an idiot chef. I think that the people complaining about their lost GPT-4o friend could use this trick with an inverse prompt if they want to bring back the toxic empathy that they love so much.”
So, Ben, do you have any thoughts on that and the GPT-4o addiction?
Well, first off, custom prompts are great. I’ve had one for ages that I always forget about. I want ChatGPT to use its own knowledge before going to search, especially because it was searching instantly, constantly, all the time. I’m like, “Don’t worry, I will verify what you say. I promise. I want you to give me information.”
I tell it to be terse and not say that it’s an AI. I do put lots of stuff in there about what I want. I stole this off Twitter ages ago; it’s not original to me. So custom prompts do help. They don’t fix the core nature of the model.
This is one of the most fascinating things. Different models have different personalities. It really is the case, and the same models from the same companies can, too. Daniel Gross was very early on this, in a super-early interview that we did with him and Nat Friedman, talking about how these companies need personality designers just like you needed UI designers.
Great UI designers are known to be incredibly rare. Creating an interface like the Apple iPhone interface is obvious now, but it was not at all obvious when it came along the first time. He put his finger on the idea that there is something about this personality aspect, and you can’t prompt that away. There’s something that’s just deep and inherent to the model.
What’s so interesting about this—I’m totally repeating a point I made the other day, so sorry to our many listeners, one of whom said that he was—but I wrote ages ago, and I couldn’t find the update. Maybe it was on Sharp Tech; I can’t remember. I wondered, if you’re an app developer building on one of these models, what’s going to happen when the model updates? Is it going to just drop right in and now your app is way more capable? Or are you going to have to do a lot of work to reconfigure yourself to the model?
It turns out it’s pretty much a drop-in, and you can actually switch models. That commoditizes the models, and you can have more competition over who’s cheaper, and so on. But so much work goes into the scaffolding around it that the model is just a piece of the puzzle.
That’s a good thing from a product-development perspective. The problem for ChatGPT is that the product is the model to a certain extent. They ran into a situation where the interface was all the same, but it was a different product for people for whom ChatGPT is not a tool but a companion.
Therapist, whatever it may be.
They changed their friend. They switched out their friend. It’s easy for us to sit here and be annoyed at the cloying nature and the affirmation. I think somewhere in my prompt I’m trying to threaten ChatGPT to stop complimenting me, and it still did it anyway. That’s how I know the prompt doesn’t work.
Totally. I have a bunch of friends who use it every day as part of their workflow, and they have to repeatedly say, “No, I just want the answers. No cloying flattery here and there.”
My question is, in terms of what Daniel was talking about, how much control is there really in terms of how these models evolve and interact with people? Do we know?
I think no one seems to know. Maybe there’s some secret sauce in here. Anthropic’s Claude 3.5, specifically 3.5 Sonnet, really had this distinct personality that a lot of people found very engaging, which you didn’t get in the larger model. Maybe the people in these labs do have a sense of how this works. I think there’s a fair bit of no one being quite sure how it works, either.
A little bit of a black box in terms of the people inside the lab, yeah. That’s part of the fun here. I’m still chasing the vibes we got from Sydney—
Oh.
—who was negging you—
My girl.
—two and a half years ago.
My girl.
Bring back that antagonistic personality. It does seem like every single LLM has that problem, though.
But that speaks to it. Sydney was GPT-4, right? There is some aspect of the post-training and alignment work they do where some of the personality is being put in.
Intentionally or unintentionally. So that was sort of Altman's response, which is, “We need to get, like, a personality selector/designer so people can have the personality that they want.” So that suggests that it is probably fairly controllable; it is a post-training process. It’s not necessarily emergent. But I guess we’ll see. If they come out with a personality tuner and it works, then I guess we’ll have our answer.
Indeed. To keep it moving, we’ll hit the mullet portion of the episode and bounce around here. We got this note from Sam regarding bubbles: “One sentiment Ben often expresses around the dot-com bubble is that everything was indeed extraordinarily useful and unlocked tremendous potential, but people went too crazy too fast with their investments.”
Sorry, I need to issue a correction.
Okay.
It was not useful, because people had slow computers and dial-up connections. What changed was widespread broadband and mobile phones. Suddenly, all these things became very useful if you could access them all the time and consistently. So the ideas were good, but the technology wasn’t there to do it. They were theoretically useful, I would say. They proved to be useful.
I think he’s referring to eventually useful—useful in retrospect. At the time, it wasn’t super rational, and Google was able to buy up all the unused fiber and make dramatic progress, but if the initial investments had been rational, it never would’ve been built by the soon-bankrupt companies.
So, Ben, I’d love some bubble commentary today. With the release of GPT-5, does it feel closer than ever to being a repeat of history? Lowercase Sam has been talking about GPT-5 potentially being AGI for a long time, but it’s only marginally better. Still, if we imagine AI remaining stagnant for a decade, there are tons of companies that can be built around the current innovations. What do you think of the bubbly circumstances that we all inhabit right now?
I’m still hung up on “marginally better.” These people with their expectations, I’m telling you. It’s better. It’s really good.
It’s better. It’s also just so much smarter for their business in terms of controlling costs, making it easier for mainstream users.
Well, so are we in a bubble? Almost certainly, yes. But the problem is that people were like, “We’re clearly in a dot-com bubble” in 1996, and there were still 40 years to go, right? So are we in a bubble, or will this end up in a bubble? Probably that’s what happens. Anything transformative that requires massive capital investments results in bubbles. That’s what happened with railways; it happened with ships back in the day; it happened with the dot-com era and electricity. Timing—you’re not right unless you get the timing right.
Yeah.
And I’m not going to be so bold as to think that I can do that. What we do see is OpenAI has real revenue numbers that are really large and growing really fast.
Mm-hmm.
Anthropic has come out of nowhere to have massive revenue numbers that are large and are growing fast.
Mm-hmm.
Microsoft is announcing incredible results on its AI business. A lot of that’s ChatGPT, but still, Google’s cloud business is growing hugely. Everyone is supply-constrained in terms of capacity to serve this. Now, are these companies experimenting and not actually finding use cases? Possibly. By the way, I had a great bot experience the other day.
Oh, boy.
My home improvement stuff: I got a package from Amazon that had nothing in it. Just an empty envelope. It used to be that Amazon was amazing. You’d go on there and say, “I got a package with nothing in it.” They’d be like, “Oh, sorry. We’ll send a new one right now.”
Then it all went to crap, and you had to jump through all these steps and blah, blah, blah. I went to Amazon, got on a chat, and said, “I got nothing in it.” It said, “Great, we’ll send you a new one.”
Really?
It was definitely not a human. They lost that human component, which was so great about Amazon a few years ago. They realized that instead of trying to parse a claim, just send them a new one and they’ll be a happy customer. But it got so convoluted as they scaled because they couldn’t serve you with humans anymore; they had a gazillion steps in front of it. That’s definitely an LLM, and a pretty dumb one sitting there. Great experience. Loved it.
Wow.
Fantastic. It’s a cheaper, better customer experience. I don’t know if it’s cheaper, but it avoids clicking through a gazillion things. It will manifest in me being a happy customer and telling this story on this podcast right now. There are so many things where we could see this being really useful right away.
Mm-hmm.
That’s in contrast to the dot-com bubble, which was so circular. Now, maybe it’s all very circular right now. Like I said, how much of Microsoft’s huge revenue is actually just ChatGPT, right?
Downstream of ChatGPT, yeah.
It’s ChatGPT driving that, actually, right? And how much of Anthropic’s revenue is just Cursor, right? This is one sort of breakthrough product that seems to be losing tons of money. So is it going to go too far? If it’s not too far right now, is it going to go too far? Yeah, almost certainly. That is going to happen. But it doesn’t mean there isn’t anything real here.
The question that I think is interesting—
Okay.
He mentioned the Google Fiber bit, because the dot-com bubble was actually 2 bubbles. It was the internet companies, which is what we mostly think and talk about, and it was the telecom companies laying broadband. They’re the ones that really lost their shirt more than anyone else. They’re the ones that tanked the economy much more than the internet companies. The dot-com losses were relatively localized. Yes, they went public and people lost money.
Where real money was lost was WorldCom and all these companies just losing tons and tons of money laying all this fiber, which Google bought. That ended up being the really important part. You had all this fiber Google built up. It undergirded the internet. You had this build-out that was maybe not economically rational, but once it’s there, it’s there.
What is the build-out today, right? It’s not GPUs.
Well, it’s all GPUs and data centers—
No. No.
But don’t they depreciate?
GPUs do depreciate too quickly. It can’t be the GPUs. It can be data centers. There was a tweet going around about some crank investor who likes to short companies, talking about earnings and complaining about their depreciation schedules.
The problem with the tweet was that he said, “Oh, they’re depreciating GPUs over 15 years.” No. What he identified was blended depreciation: data centers depreciate over 30 years, and GPUs depreciate in 5 years. Now, 5 years still might be too long, to be honest, for how long they’re used. But actually, if you construct a building and wire it all up, you’re going to use that with multiple generations of GPUs. You should not be depreciating your data center on a 5-year depreciation schedule, because you’re going to use it for much longer than that. So is that the long-lived asset?
Could it be power?
It doesn’t seem tangible. Bingo. That’s power. If we want this bubble to be a bubble that matters, like the dot-com bubble and the telecom bubble—a bubble that mattered for spreading the internet—what we need is insane build-out of power. Everyone builds it out, goes bankrupt, but guess what? We now have massive power. We get excess power.
Yeah.
And then you think about that. If there is a bubble, this would probably be bad, to be clear. We’ll probably get a recession. Everything would be terrible. But a world of excess power, just like in 2000 we had a world of excess fiber, created the conditions for what followed.
A world of excess power would be an incredible one. It’d be very painful to get through, but this will be a productive bubble if we get power.
And I’m not clear on how much progress we’re making on that front.
That’s very depressing. I was going to write about this, wrestle with the inflation statistics on Stratechery. Becoming an economics reporter—very dangerous terrain over the last couple of weeks. Energy was down, which was a big reason why inflation looked lower. A lot of stuff was actually more expensive, but energy was way down.
But energy was down because gas is way down, like 10% or something like that. Electricity is up, and it's been up and up and up, and it's up 10%, 12%, 15% from last year.
Yeah, it's a big problem.
Guess what? You have Meta out here, you have OpenAI, Stargate, whatever, you have Microsoft. At this point, they will buy electricity whatever the price is, and utilities will sell it to them, and that means everyone else is going to pay more, because it has to be fixed by building new power. And there's all this stuff about Nvidia chips, like chip controls—all this is a total sideshow.
Mm-hmm.
Are we going to have enough power? China's going to have enough power. Are we?
Yeah.
And it would be a great thing, and one of the most productive bubbles in history, if we can get a bunch of people to go bankrupt building new power generation.
Well, I'm going to hope for the best on that front. Again, it's early in the Trump administration, but over the last 6 months there hasn't been that much movement on power.
I mean, that was certainly one of the hopes, right?
Yeah, exactly.
They would really drive deregulation around this particular area and push things forward.
There you go. Well, there's your bubble commentary, Ben. We'll keep it moving.
Mark says, “Ben, listening to last week's Sharp Tech and Dithering, and your opinion that Apple should buy a model company like Anthropic or Mistral, I understood that to be the advice you'd give Apple's team. But do you also think that's best for the market broadly? That is, if you were speaking for America broadly, or even just the US stock market, what outcome do you think is best?
“If I recall correctly, you hold a broad index to avoid conflict of interest. Given that lens, what do you want? Should Apple buy someone, or is it better if Apple stays light on AI, keeps making money and distributing it back to shareholders as it slowly fades to irrelevance and allows OpenAI, Google, or someone else to find the next device and model? You've also written about Apple helping build China's industrial base. Does that change the answer if we care about American outcomes?
“Would it be better if a new Western-aligned firm found the next device in an environment that actively encourages Western manufacturing versus Apple with its continued reliance on China? Does the market and America do better long term if Apple takes a step back? What do you think?”
I say this with all sincerity.
Okay.
I don't think about my stock portfolio at all when writing. Actually, when you read this, I'm like, “What do I want?”
That's why I thought it was kind of fun, because it sort of—
Yeah.
It removes you from the role—
I know. Unfortunately, this has probably cost me a tremendous amount of money by not investing my book. But no, it was sort of interesting to think about. I love the China one, to start at the end; it was quite interesting. Yeah, probably good for America if Apple was not so dominant, given the reliance on China, and if we had someone that grew up in the US at a similar scale. Now, is that possible?
I don't think that's possible at this point.
That's another set of questions, to say the least.
Yeah. Google with Samsung would be our best bet for a Western-aligned manufacturing base.
Yeah, by and large, my overall philosophy has long been—and James Allworth and I used to have long debates about this on Exponent—that I don't care about companies being disrupted. I don't care about stock. I like stock buybacks, because then there's always an ongoing complaint, right? They're not investing; they're doing stock buybacks. I'm like, that's great. Give the money back to investors who can go put it in a startup and actually get something—
Right.
—much better geared for innovation, et cetera. And I think that would probably be my default answer. My advice for Apple is different from what's great generally speaking. It's a good question in that regard. That said, there is a bit of a great-man-of-history debate, you know?
Mm-hmm.
In tech, Steve Jobs was a great man of tech. He changed the course. Phones will look different, computers look different. The GUI existed, right? He refined it, but there was a very clear direction that everyone sort of followed, right?
Yeah.
And to what extent is Apple that company? Things are different because Apple exists, because Apple approaches things in particular ways. Everyone copies them for a reason. Is it actually good for the world that Apple continues to exist? Because there's something unique and special about them that you're not going to just get from general startups.
Yeah.
I don't know. I think it's interesting—it's funny how these broad-based philosophical debates can be applied to a question like this. You can make a generalist argument that companies should die, but there is something about Apple that's unique.
Mm-hmm.
They push things forward in a magical way that's superior to the typical consumer electronics experience.
They have a 40-year track record of doing it, right?
Yeah.
Do we want that to disappear? I don't know. I think that's the question.
My counter would be that's why I was excited for the Vision Pro, and the Vision Pro probably isn't the form factor that will popularize VR the way Steve Jobs popularized—
Yeah.
—the GUI.
Still an awesome device, though. I think the mistake Apple makes when Apple gets in trouble is that they underappreciate the extent to which they do benefit from others, right?
Mm-hmm.
So the GUI on the PC arrived just in time, and it was invented by Xerox PARC, famously, right? The iPhone sort of arrived right on time. It rode on top of the iPod. The iPod benefited from piracy, music storage, and the internet—all these sorts of things.
Apple—and this is why the great-man-of-history debate is so interesting. Do the trends of history matter? Yes, of course. There's a reason why multiple people invented calculus at the same time. But the answer I've settled on in this debate is: Is it A or is it B? My answer is yes.
Mm-hmm.
The trends matter, and also people make stuff happen. I think Apple at its best is a company that makes things happen at a time when the world has trended to a place where it's useful for them to make things happen. The Vision Pro is arguably Apple trying to make something happen when the world's not ready. Maybe they looked at Meta, who made the same mistake on steroids, and it's like, well, maybe you should be taking your cues from the right time and place, from those guys.
Yeah. Well, I think Apple had good grand ambitions and good impulses on the Vision Pro. They just—
The Vision Pro remains awesome to use.
Yeah.
Every time I put it on, it's great. I also haven't put it on in weeks, so.
No, but I don't mean to denigrate it, because I think some of what was driving that product should be celebrated.
No, but that's the whole point.
And so hopefully more of that spirit is kept alive out there.
All the Apple parts of the Vision Pro are awesome.
Yeah.
What's missing is everything around it—a reason to use it—which has to come from other folks. I blame Apple for their attitude toward developers, and I think that's part of it. But there's also a bit where the need exists. There's a hole in the market and you fill it. Where's the hole in the market for the Vision Pro?
Yeah. I will say there was nothing magical about the googly eyes that they had on the outside of the Vision Pro.
Yeah, no, it's dope.
So we do have to dock Apple some magic points on that front.
Sidharth says, “I've been intrigued by Gen Z's sudden nostalgia for old tech, especially BlackBerry devices. Against the backdrop of recent Sharp Tech and dithering discussions, and these are discussions about thin clients, I wonder if we're headed toward a future dominated by lightweight devices running AI operating systems, à la Her.
“Could a company like BlackBerry realistically reenter the consumer device market? Imagine a modern, low-cost BlackBerry designed as a stylish, nostalgia-rich thin client and hardware interface for AI, aimed squarely at the Gen Z market.”
Could it be the right timing for this Canadian powerhouse to make a consumer comeback? Ben, any thoughts on the prospects of a BlackBerry comeback over the next 10 years?
No.
No, not really?
7. BlackBerry's Thin Client Future
A world of thin clients powered by this AI cloud is a world of commodity hardware, right?
Yeah.
Is there an opportunity to build something cool there? Sure. There are cool Chromebooks. But BlackBerry was a fairly self-contained ecosystem.
Mm-hmm.
You’re not going to get a company like that. Whatever surrounds the core is inevitably commoditized. It’s not going to be highly differentiated. If the core is this cloud-based AI Her, whatever, the devices will commoditize around it, and that’s just the way it is. There’s generally not room for a highly differentiated product in another space.
Now, you can still have—there are TVs, right? There are high-end TVs and low-end TVs. You could have high-end devices and low-end devices. The high-end TVs don’t make any money either, right? It’s still a highly competitive market because you’re all relying on the same content.
Yeah.
You’d be relying on the same AI. So maybe it’ll come back as a nostalgia brand. I know point-and-shoots are a big thing these days. There’s lots of this sort of stuff going on.
Mm-hmm.
I don’t know if it makes me feel old. It makes me feel—I’m not sure what it is, but good for the kids.
Well, I enjoy it. It’s a throwback to the BlackBerry era. I enjoyed the BlackBerry movie a couple of years ago. I was an enthusiastic BlackBerry user.
It was a good movie. I enjoyed it as well.
It sounds like if we do enter that commoditized world, there would be room for some crappy nostalgia-bait, thin-client, low-end product. I want Sidharth to keep his dreams pretty modest in terms of a BlackBerry comeback.
Here’s what I would say on an encouraging point: When the phone became the core of the value chain, all devices basically disappeared and got subsumed into the phone, right? And then you had a new accessory market around the phone, like cases and chargers. All this stuff around the phone is a mass commodity market, but everyone had basically the same device.
Now, of course, there’s differentiation with the iPhone and different Android models, but I wrote an article ages ago with a picture of my camera and my calculator and all this sort of stuff, and how it went into the phone. Those are now apps on a phone.
Yep.
One thing that could be exciting for devices—maybe we’ll get the return of device blogs—is if this AI is all in the cloud and can manifest through an earpiece, through glasses, or through whatever. We would actually expect to see an explosion of devices because the devices are accessories.
Mm-hmm.
The accessory market can be fun and cool. It’s not going to be super profitable because anyone can make one. That’s the whole thing. It’s a modular, commoditized aspect around the core. I think that actually does sound pretty cool, and maybe we could have a keyboard-based AI device.
I was going to say, it isn’t the craziest thing in the world to have a device for text messaging and emailing, and then everything else is handled in the cloud. All your apps are in the cloud. That sounds appealing to me as a consumer.
Here’s the question, though: When’s the last time you typed on a BlackBerry keyboard?
It’s been a while, but I was very, very good on a BlackBerry keyboard 15 years ago, so I feel like I could probably pick it up again.
I 100% see you buying this device and abandoning it plus or minus 3 and a half days.
Oh, boy. Look, I thought about it because I read this note, and I was like, “Could I actually live without the apps that I have become addicted to on the iPhone, and would I be comfortable if those were all just ethereal, in the cloud, voice-activated?”
You know, I will say, I think you’re a pretty effective and talented emoji user. You don’t overdo it. You choose the right one at the right time.
Mm. I am quite good, yeah.
Are you going to live without an emoji keyboard?
Well, I would guess that BlackBerry could integrate that. I think they had emojis in BBM way back when.
We’ll see. I’m more likely to abandon the iPhone because I was also thinking about going with a Samsung Galaxy at some point for the sake of the show. I’m more likely to take the leap for a thin client that is a really simple interface, and this is 7 years down the line because all this is a long ways off. Or I guess it’s not that far off if we’re talking 7 years.
That is more appealing to me than leaving the iPhone ecosystem for Google at this point. So sorry to disappoint the Google fans in the audience.
Oh.
That is disappointing.
We’ll just have to hope Google Glass is really terrible, and then you’ll bail. Does this device have to be wired?
Wired for what? For headphones?
I don’t know. You’re the EarPods-over-AirPods guy.
If there’s no headphone jack in the nostalgia-rich BlackBerry device that we’re dreaming about in the future, I’m going to be very pissed off. We’re bringing back BlackBerrys. We’re also bringing back the headphone jack at long last.
You do you.
Dead silence and disappointment. Great stuff.
8. The Bell Labs Monopoly Tradeoff
And speaking of disappointment, 2 final notes here. This is Bill. He says, “I’ve heard Ben talk wistfully about the AT&T monopoly and the good old days when Bell Labs invented cool stuff for the good of humanity. Friday’s episode takes the cake, though, with Google’s Genie 3 as the catalyst for a new level of lamenting the end of the AT&T monopoly.
“There’s a long list of technologies and products that AT&T and Bell Labs suppressed or delayed the release of. 1. Tape-based and early solid-state answering machines. 2. Bell Labs’ Picturephone, first demonstrated at the 1964 World’s Fair. 3. Magnetic recording and home media. 4. Early modems and data networking, late 1940s, early 1950s.” He says, “Mobile telephony is number 5. And 6. The Unix OS.”
Now, Ben, each of these categories came with a description that I’ve omitted from the outline—
Oh, I was going to add—
—just for brevity’s sake.
My own. Like, for example, answering machines. Terrible. I’m glad they delayed it. Who wants an answering machine? Picturephone, also terrible. Who wants it? They had delayed it. No, sorry, I digress. Continue.
I just think it’s interesting that Bill lists a bunch of stuff that Bell Labs invented. So is he making my argument or his argument?
They were early.
First off, Bill, one of my longstanding points around Bell Labs is that it is one of the accidental great antitrust cases in history. There was a settlement where they had to basically open-source their patents, and anyone could use them. Everyone was mad about it. They were like, “You didn’t punish AT&T enough.”
It basically seeded the entire computer industry. It continually cracks me up that the 2 greatest antitrust cases were accidents, which is the AT&T one and the IBM one, where IBM unilaterally separated hardware and software to avoid an antitrust case. That came down anyway. The case had no impact. The split had a huge impact.
Again, though, I go back to my first point: They invented all this. There’s this magical world where this stuff got invented by someone else and then was dispersed way more broadly because AT&T didn’t limit it. I’m not arguing monopolies are good. I’m saying there are benefits that come from a company basically becoming the outsourced R&D arm of the U.S. government: “Okay, we’ll let you have your long-distance money, but spend a lot of money inventing stuff and open-source it. Make it broadly available.”
Invent the technology industry for us.
Right. You can’t deny the stuff did get invented. It did get—what’s the word I’m looking for?—not dispersed, diffused.
Yeah.
And a lot of it is incredible. He lists the Unix OS as a detriment here.
It was diffused, though, because of the antitrust action, to be fair.
No, but the point—the problem here is, and my problem with the antitrust movement in general, is this absolutist view and approach that can't even grant a bit of legitimacy or doubt to their position. Stuff just magically invents and gets out there, and all monopolies are the government's fault and nothing bad will ever happen if you mess them up. The point about Bell Labs: of course the AT&T monopoly is bad. Of course there were issues with it. I'm not denying those points. It's also okay, Bill, to admit some good stuff happened, and it's not for sure that that good stuff would have happened, or happened on the timeline it did, without that.
Hmm.
They invested tons of money into Bell Labs. They hired the best people in the world who were able to do whatever they wanted to. And I have a hard time believing that all the stuff they did and invented would have happened when it did, at the time it did, without that largesse, and that largesse was 100% downstream from a monopoly. Does that say I'm defending monopolies up and down? No. But if you can't even grant this bit of grace—that maybe there were a couple of good things that happened—it's kind of hard to take the monopoly argument seriously. And, sorry to label you, Bill, as the stand-in for the whole movement, but that is kind of the whole thing that drives me nuts.
Yeah. Well, I think Bill is just adding some footnotes to your comment last week, which was partly tongue-in-cheek in terms of the Bell Labs monopoly. But there are real benefits as well.
The AT&T monopoly was not the good old days and never was. I think there were some good days. I feel like there were some good days.
In some ways, it was the good old days. Yeah. If we ever release merch, my one request is that we do release, like, an anti-anti-monopolist shirt. Because every time you say that on the podcast, it cracks me up to no end, even as someone who’s more of an anti-monopolist. The way I think about the ecosystem, though, is—not to get too political, but actually, antitrust is unavoidably political, no matter what you're talking about.
That's exactly true. That's realism.
When you look out at the states that function best, I think the states that are purple function better than the one-party states, whether you're looking at Democrats or Republicans. When there's some real tension in the system, those states tend to produce the best results across a variety of categories. And purple is how I want the antitrust process to look, where both sides make some good points and good arguments, and there should be real tension there. And I think for a long time there wasn't.
Now there's more tension, and everything is functioning properly as far as I'm concerned. So we'll see what the judge rules on Google's fate later this month. I think that ruling is coming down soon.
Someone proposed, what, making Google open-source the patents. It worked one time.
Absolutely. Let's do it. And speaking of Google, Kevin says—
You know what? Leave Google alone.
Let Google cook.
Let Google cook.
It was “leave Google alone and let Google cook.” You've turned into a Russell Westbrook fan. Let Russell Westbrook be Russell Westbrook. Let Google be weird.
9. Google's Slime Mold Problem
Kevin says, “In the middle of last week's Sharp Tech, you guys called Google an amorphous blob. Well, there's a famous internal memo at Google titled Why Everything Is So Hard at Google, which compares Google to a slime mold. Business Insider wrote about it a few years ago and said, ‘The most memorable part of the memo compares Google's bottom-up organizational structure to a slime mold, highlighting how both Google and a slime mold can work independently but still come together to solve complex problems. It's a strength and a weakness. A culture that prizes autonomy can accomplish big things, but the larger the organization grows, the slower things get. As each fiefdom within Google operates on its own, Google engages in messy behavior that can be, quote, “hard to predict.”’”
So there you go, Ben.
Yeah.
You were right on the money.
You know what's long-lasting and hard to kill?
What's that?
Mold.
That's true.
I'm not a fan, though. I'm not sure this is endearing me to Google. I do remember that. Someone had a thing in here that you didn't read about, like Ben-ception or something, where someone was repeating my argument.
Mm-hmm.
That was clearly me repeating this memo. I remember it now that the emailer mentioned it. I pretended it was an original insight, but it was not.
But accurate.
Nevertheless, an accurate depiction. Mold, I guess, could be how monopolies go wrong. And so that's sort of the danger of entrenched companies that could just hang out and invent cool stuff and never release it as products.
But on that note, I hope everybody has a great weekend. Ben, I hope you have a great weekend. People can continue emailing us, email@sharptech.fm, and I will talk to you on the other side. Hope you touch some grass this weekend.
Thank you. I'll talk to you later as well.