> ## Content Index > Fetch the complete content index at: https://spyglass.org/llms.txt > Use this file to discover other available public pages before exploring further. # Signal: Lego Heads 📧 - URL: https://spyglass.org/signal-lego-heads/ - Published: 2025-10-20T13:24:02.000Z - Updated: 2025-10-20T13:25:46.000Z - Description: Apple & F1 • Karpathy on Dwarkesh Pod • Xitter Rethinks Links • Meta's Private Capital AI Play • iPhone Air Sales • Cue's Streaming Content Critique - Author: M.G. Siegler - Tags: newsletter, signal Back on the road again this week (have I mentioned how many breaks children get from school in Europe?), but as always will be writing here and there, just more sporadically. --- ### I Wrote... [The Checkered Flag is Out, Apple Finally Takes F1 RightsAnd all Apple TV subscribers will benefit!![](https://storage.ghost.io/c/af/ca/afcaa655-46e2-45b8-889a-2881de5cce69/content/images/icon/Spyglass-Rings-----Multi-448.png)SpyglassM.G. Siegler![](https://storage.ghost.io/c/af/ca/afcaa655-46e2-45b8-889a-2881de5cce69/content/images/thumbnail/ChatGPT-Image-Oct-17--2025--05_06_50-PM-1.png)](https://spyglass.org/apple-f1-rights/) --- ### Thoughts On... **🗣️** [**Andrej Karpathy on the Dwarkesh Podcast**](https://www.dwarkesh.com/p/andrej-karpathy?ref=spyglass.org) – The [headlines](https://www.techmeme.com/251017/p26?ref=spyglass.org#a251017p26) are zeroing in on his poo-poo-ing of the imminent AGI hype, but there's a lot more in here (as you'd hope from a two and a half hour podcast!). Even just the first 30 minutes or so is worth your time as Karpathy lays out why he feels like all the comparisons of LLMs to the brain are wrong – hint: [it's evolution, baby!](https://www.youtube.com/watch?v=aDaOgu2CQtI&ref=spyglass.org) – and why what we're actually building with the current AI systems may be more akin to ghosts than actual humans (they're echoes of us). And why dreaming (or the equivalent) may be a missing ingredient, in both flushing unneeded memory and fully committing other core learnings to the repository. It's almost more philosophical than technical, but we get to that as well. Such as when Karpathy breaks down [his recent "nanochat" project](https://x.com/karpathy/status/1977755427569111362?ref=spyglass.org) and conveys just how much vibe coding still mainly sucks for all but the most rudimentary tasks (which is still valuable, as he notes). Mainly, it's somewhat amazing how eloquent while being erudite Karpathy is compared to some of his peers in the field. **\[**[**Dwarkesh**](https://www.dwarkesh.com/p/andrej-karpathy?ref=spyglass.org)**\]** **🧠** [**Xitter Rethinks Links**](https://www.engadget.com/x-is-testing-a-new-way-of-opening-links-in-posts-to-improve-engagement-211210520.html?ref=spyglass.org) – To be clear, it's [not entirely clear](https://x.com/nikitabier/status/1979994223224209709?ref=spyglass.org) that this signals a change in the way Xitter handles the ranking/surfacing of posts with links in their algorithm, but it at least *suggests* that they might go back to making such posts first-class citizens again. I know I'm not alone in thinking that just [practically speaking](https://spyglass.org/a-social-media-nightmare/), this was [the worst change](https://www.threads.com/@mgsiegler/post/C1h9fzHI3AP?ref=spyglass.org) of the Elon-era. Twitter used to be *the* place to find and share links and just scrolling right now to try to see if I can test this new functionality, I'm hard-pressed to find *a single tweet with a link shared* in the main tweet (with most being relegated to a second tweet, which remains just farcical hoop-jumping bullshit). Obviously, the company believes this was a good change in an attempt to move beyond Twitter's roots and into [the "everything app"](https://spyglass.org/x-marks-xais-spot/), but this change – help us, Nikita Bier, you're our only hope – may be an admission that it hasn't worked. Twitter, even as Xitter, [remains](https://spyglass.org/twitter-x-cnn/) the real-time news hub of the world. Just with a lot more spam and porn and a lot fewer links these days. Here's hoping... **\[**[**Engadget**](https://www.engadget.com/x-is-testing-a-new-way-of-opening-links-in-posts-to-improve-engagement-211210520.html?ref=spyglass.org)**\]** **💰** [**Meta's Private Capital AI Play**](https://www.bloomberg.com/news/articles/2025-10-16/blue-owl-seals-largest-private-capital-deal-for-meta-s-ai-growth?ref=spyglass.org)– It's not just [OpenAI](https://spyglass.org/openai-rising-tide-strategy/) and [xAI](https://spyglass.org/signal-the-perpetual-money-machine/#:~:text=%5BInformation%20%F0%9F%94%92%5D-,%F0%9F%92%B0%20xAI%20Raising%20%2420B,-%E2%80%93%20Speaking%20of%20Elon) that are figuring out ["new ideas"](https://spyglass.org/ai-data-center-build-outs/) for financing the billions needed for data center build outs. The difference, of course, is that unlike those two, Meta is *wildly profitable*. Still, this private capital SPV structure allows them to raise massive amounts of debt without it hitting their own balance sheet. And it allows Blue Owl – [also the co-owner of the first 'Stargate' data center](https://spyglass.org/the-stargate-data-center-layer-cake/) in Texas – to own most of Meta's 'Hyperion' data center in Louisiana (with Pimco as the anchor lender). And presumably the same model will be used for the [just-announced](https://www.bloomberg.com/news/articles/2025-10-15/meta-building-new-gigawatt-sized-data-center-in-texas?ref=spyglass.org) new 1GW data center Meta is building in Texas too, alongside 'Prometheus' in Ohio – no word on the Greek God name for the Texas one yet. **\[**[**Bloomberg**](https://www.bloomberg.com/news/articles/2025-10-16/blue-owl-seals-largest-private-capital-deal-for-meta-s-ai-growth?ref=spyglass.org) **🔒\]** 📲 [**The iPhone Air Up There (And Down There)**](https://9to5mac.com/2025/10/17/iphone-air-sells-out-almost-instantly-in-china/?ref=spyglass.org) – A lot of conflicting reports as to just how popular (or not popular) the iPhone Air is thus far. While it [apparently sold out in China](https://www.scmp.com/tech/big-tech/article/3329336/apples-iphone-air-sells-out-minutes-china-after-ceo-tim-cooks-visit?ref=spyglass.org) after its ([delayed](https://spyglass.org/signal-openais-pbc-race-is-on/?ref=spyglass-signal-newsletter#:~:text=So%20much%20for%20those%20iPhone%20Air%20pre%2Dorders%20in%20China%2C%20as%20the%20device%20is%20now%20clearly%20delayed%20while%20all%20the%20carriers%20%28and%20the%20government%29%20scramble%20to%20ensure%20they%20can%20handle%20the%20eSIM%20element%20%28at%20first%20only%20China%20Unicom%20was%20on%20board%2C%20which%20presumably%20the%20other%20carriers%20didn%27t%20love%29.%20%5BSCMP%5D)) release, in the US and other Western markets, Apple may be cutting production already. Anecdotally, here in London, the Air remains widely available in stores while the iPhone 17 Pro is constrained. But even more interesting, the iPhone 17 "regular" seems to be the *most* constrained. The reports seem to back up that it's [selling far better than the iPhone 16](https://www.ft.com/content/be3f6853-286b-4296-9cef-fd0a9da624f8?ref=spyglass.org). It's possible Apple simply underestimated demand for the base model? Maybe people dig the colors? If the above is true, curious how Apple handles such data. Are [future iPhone Airs](https://spyglass.org/apple-iphone-17-event/) aimed directly at China while iPhone "Regular" is more targeted at the West? Does it alter what was clearly the plan to slowly morph the Air into the "Regular" with iteration? Was I right in [my pre-launch thinking](https://spyglass.org/the-3-iphone-problem/) as to [why the Air was a gamble for Apple](https://spyglass.org/signal-iphone-unveil-week/#:~:text=%F0%9F%93%B2%20Apple%27s%20%27iPhone%20Air%27%20Gamble)? **\[**[**9to5Mac**](https://9to5mac.com/2025/10/17/iphone-air-sells-out-almost-instantly-in-china/?ref=spyglass.org)**\]** **💩** [**Will AI Products Get Enshitty?**](https://www.wired.com/story/can-ai-escape-enshittification-trap/?ref=spyglass.org) – Steven Levy takes the publication of [the book](https://www.amazon.co.uk/Enshittification-Everything-Suddenly-Worse-About/dp/1836742223?ref=spyglass.org) on "[enshittification](https://en.wikipedia.org/wiki/Enshittification?ref=spyglass.org)" to ponder the question author Cory Doctorow doesn't address (as it's still too new): how will this play out with AI? I mean, *obviously* it will play out the same as it has with every other product: the incentives will be too strong for this *not* to happen. Again, that's always the case (and why this happens in general, obviously). But I also think Doctorow (when Levy asked him) is right that it could happen *faster* with AI because of both the "black box" nature of these services and the [insane amount of money already flowing](https://spyglass.org/openai-rising-tide-strategy/) through the systems to [build them](https://spyglass.org/ai-data-center-build-outs/) and [make them work](https://spyglass.org/openai-spend-cloud-chips/). Which is to say: enjoy the (subsidized) AI products we have right now. They sadly won't last. As least not in the same surprising and delightful way that we now enjoy. **\[**[**Wired**](https://www.wired.com/story/can-ai-escape-enshittification-trap/?ref=spyglass.org) **🔒\]** --- ### I Quote... > "The problem though, is we all know we can all watch everything, but it's very hard to find. You have to sign up for 1,200 subscriptions around them." – **Eddy Cue**, [talking about](https://finance.yahoo.com/news/apples-top-exec-says-things-033114483.html?ref=spyglass.org) the current problem with streaming content. Not mentioned, [you also have to *pay*](https://spyglass.org/expensive-streaming-bundle/) for those "1,200" subscriptions. All of this is why the players, [including now Apple](https://spyglass.org/signal-how-you-like-them-apples/#:~:text=%F0%9F%8D%8E%F0%9F%A6%9A%20Apple%20TV%20%2B%20Peacock), are [re-bundling](https://spyglass.org/espn-fox-one-sports-bundle/). And that will continue. But [the real answer is an "iTunes" for the space](https://spyglass.org/itunes-for-streaming/) – a unifier that just makes it all work, seamlessly. Apple, led by Cue, has been *trying* to do that for years already, but others – [notably](https://spyglass.org/netflix-apple-tv/), [Netflix](https://spyglass.org/apple-tv-netflix-streaming/) – won't play ball. In part because they know what Apple wants to do: they saw it first hand with iTunes! What's great for consumers is less great for them... --- ### Asides... - Amidst the [AI talent brain drain](https://spyglass.org/apple-ai-talent/), there are whispers inside Apple that **iOS 26.4** – aka the "**New Siri**" – *still* isn't going to be [up to snuff](https://spyglass.org/apple-ai-siri/)... **\[**[**Bloomberg**](https://www.bloomberg.com/news/newsletters/2025-10-19/apple-m5-ipad-pro-vision-pro-macbook-pro-impressions-review-touch-screen-mac-mgxnn1w8?cmpid=BBD101925%5FPOWERON&utm%5Fmedium=email&utm%5Fsource=newsletter&utm%5Fterm=251019&utm%5Fcampaign=poweron) **🔒\]** - Other whispers suggest a team within Apple is quietly working on [the deep](https://spyglass.org/apple-siri-gemini-ai/) **iOS** [integration](https://spyglass.org/apple-and-google-are-so-back/) for **Gemini** (and Gemini 3 remains on track for December)... **\[**[**Sources**](https://sources.news/p/google-readies-gemini-3-perplexity?ref=spyglass.org)**\]** - Speaking of **Xitter**, they're also launching a **marketplace** for **inactive** **Twitter handles**, but you have to pay the top tier prices to access them. **\[**[**Verge**](https://www.theverge.com/news/802474/x-is-launching-a-marketplace-for-inactive-handles?ref=spyglass.org)**\]** - Oh, and they're working on a way to essentially **verify** **accounts** without using the method they had and destroyed for verification. **\[**[**TechCrunch**](https://techcrunch.com/2025/10/15/x-plans-to-show-more-information-about-user-profiles-to-help-improve-trust/?ref=spyglass.org)**\]** - **Nintendo** launched with solid **Switch 2** inventory and saw [record sales](https://spyglass.org/nintendo-switch-2-launch/) as a result, to the point where they continue to ramp supply. Feels like Playstation and Xbox could learn something there ([if Microsoft even remains in this business](https://spyglass.org/signal-apple-tv-for-you/#:~:text=%F0%9F%A7%A0%20The%20Xbox%20%24360%20Era)). **\[**[**Bloomberg**](https://www.bloomberg.com/news/articles/2025-10-17/nintendo-switch-2-production-plan-25-million-units-by-march-2026?ref=spyglass.org) **🔒\]** - Seems like **Ilya Sutskever's** OpenAI stake and whatever the "**Brockman Memo**" is may become bigger news soon with a court ordering both into the open. This is all from [the Elon lawsuit](https://spyglass.org/elon-musk-openai-chaos-bid/), of course. **\[**[**Information**](https://www.theinformation.com/briefings/sutskever-disclose-brockman-memo-openai-stake-musk-suit?rc=lsmcir&ref=spyglass.org) **🔒\]** - All things can be true: ***One Battle After Another*** can be [fantastic](https://spyglass.org/signal-one-subversion-after-another/) and the biggest movie in P.T. Anderson's illustrious career, and also [a big bomb](https://spyglass.org/one-battle-after-another-box-office/) at the **box office** (due to its budget), with $100M in projected losses. **\[**[**Variety**](https://variety.com/2025/film/box-office/one-battle-after-another-lose-100-million-dollars-theaters-1236552914/?ref=spyglass.org)**\]** --- ### I Spy... [Police raided the home](https://www.nytimes.com/2025/10/18/us/lego-theft-california-arrest.html?ref=spyglass.org) of someone suspected of stealing tens of thousands of Lego pieces and found... Lego figurines with their heads removed so they could be *organized by facial expression*. OCD much, (alleged) Lego thief guy? [![](https://storage.ghost.io/c/af/ca/afcaa655-46e2-45b8-889a-2881de5cce69/content/images/2025/10/17xp-lego-jumbo.webp)](https://www.nytimes.com/2025/10/18/us/lego-theft-california-arrest.html?ref=spyglass.org)