Dispatch 032: The PE Playbook for America

The dinners are one thing. The payments are another. Nothing illegal, of course. But a general feeling of ick around the whole charade and parade of people down south. Everyone hopes for the best. But this all feels very different...

America Has Been Acquired

This has felt less like an election and more like a deal…

SpyglassM.G. Siegler

I Think...

📲 Apple Reveals Most Downloaded Apps and Games of 2024 – A few things worth noting here: amongst the top free iPhone apps, 3 are owned by Google (Google, YouTube, Gmail), 3 by Meta (Threads, Instagram, WhatsApp), 2 by ByteDance (TikTok, CapCut) – which leaves just 2 other apps: ChatGPT at #4 and Temu at #1. If TikTok really goes away, is CapCut safe? What about Temu, also Chinese-owned, of course? Threads clearly had a massive year at #2 – no Bluesky here, but it undoubtedly came on too late in the year. Xitter? They're doing great, they say, but most third-party data seems to suggest otherwise. Interesting to see the Google app so high up here. It's a decent enough app, but presumably people are downloading it mainly because they search for 'Google' when setting up a new phone? Or maybe Gemini was driving it, as this app was the only way to access it until the recently released Gemini stand-alone app for iOS... Top free iPad apps are mainly streaming services, as you'd expect, but funny to see Calculator (Pad Edition) on there, even with the newly built-in iPadOS Calculator app (perhaps that changes next year with a full year on the market). Also, Chrome continues to do much better (relatively speaking) on iPad versus iPhone. Maybe because people are trying to use it more like a computer? [MacRumors]

🎅 ‘Red One’ Lands Amazon’s Biggest Streaming Debut – 50M worldwide viewers in the first four days of release on Prime Video sounds impressive, but Amazon gives absolutely no context for what constitutes a "viewer" – i.e. how long do they have to watch for in order to be counted, etc. Regardless, #1 is #1 and Prime Video has a lot of content, so it's likely big. And it speaks to their strategy here – Red One debuted in 3,000 screens in the U.S. and 3,300 worldwide, taking in $175M. That number is not impressive given that the movie apparently cost $250M before marketing. That's a pretty big flop. But if you want to consider that run as marketing leading up to a massive opening on Prime Video... maybe it's a little less of a flop. You'll note Jennifer Salke's (head of Amazon MGM Studios) quote is all about this two-pronged approach. That's obviously intentional as it stands in stark contrast to Netflix's approach (for now). And with Apple pulling back from theatrical in a major (and majorly stupid) way, Amazon sort of stands-alone (well, with Disney) as the key partner for talent that wants both an opening in theaters and wide distribution on streaming. Still, $250M+ is a lot to spend. Especially since the movie was awfully mediocre (we were one of those 50M "viewers" this past weekend – again, no idea how they count that since my family watched). Amazon is sort of in Netflix territory here – as is Chris Evans (of course, that's nothing new). The Rock just really needs to stop making movies titled Red ____. I guess it surpassed the decidedly awful Road House remake though. So there's that. [THR]

📺 YouTube Will Now Let Creators Opt-In to Third-Party AI Training – Sort of surprised by how "open" Google is being here. While it's off by default, they're now letting creators choose if they wish to let their videos be used to train models made by the likes of: AI21 Labs, Adobe, Amazon, Anthropic, Apple, ByteDance, Cohere, IBM, Meta, Microsoft, Nvidia, OpenAI, Perplexity, Pika Labs, Runway, Stability AI, and xAI – or even anyone else, if they want to keep it fully open. While many of these seemingly already used YouTube videos in the early training of their models, it's just interesting that Google is keeping the doors open to such valuable data if a creator chooses. Presumably that's to keep both creators happy (as this might open the door for a new compensation model down the line) and perhaps to keep regulators at bay (imagine the optics if Google was the only company able to use YouTube data for training purposes – we'd be getting calls to spin-off YouTube instead of Chrome). Still, xAI has moved to exclusive use of their own proprietary data from Xitter. Meta uses their own tangential services data. Etc. Good for Google, but is it... good for Google? [TechCrunch]

👀 Google DeepMind Unveils a New Video Model to Rival Sora – Tangential to the above, Google would clearly – clearly – like you to know that 'Veo 2' is more powerful and capable than Sora. That's great. The problem remains that no one seems to have any idea how to actually try it out. I worked there for over a decade and I can't begin to tell you how you access it. Googling it isn't even straightforward. You're taken to a site to sign up, not to access Veo 2 directly, but to access 'VideoFX' which is another tool/service that's hosted on a Google DeepMind sub-site. Clicking on that takes you to yet another sub-site, this time for Google Labs, before it auto-redirects you to yet another sub-site where you sign-in to get a "sneak peek" at Veo 2. Clicking that prompts you to log in to AI Test Kitchen, yet another product/service. Once you do all that, you're prompted to sign up for a waitlist. I get that Google is taking a cautious approach to this – as they always do – but if you want to tout your advances to the world versus your rivals, it's not really honest to obfuscate their usage to such extremes. Especially when those rivals have a simple website to showcase their wares. But hey, at least the new 'Imagen 3' is easy to find. You simply use ImageFX. Where is that? Nobody knows. Have I mentioned that none of these are Gemini? [TechCrunch]

I Link...

• Kudos to United for quickly jumping on the new AirTag sharing location feature to be able to help find lost luggage. I don't travel anywhere without AirTags in our bags and it saved me once in a similar vein already. [Verge]
• One of Intel's new (interim) co-CEOs said at a conference that the return rate for Qualcomm-based PCs are high, so Qualcomm responded that such rates are "within industry norms" (which is obviously and intentionally vague) and goes out of their way to add that Qualcomm "expects 30 percent to 50 percent of laptops to move to non-x86 platforms" in the next five years. Fun. [CRN]
• In other great moments in PR, Masa Son announces with Donald Trump that he will invest $100B in the US – money which includes some already-made investments, such as the recent $1.5B in OpenAI, and obviously ones he was going to do anyway. Still, it may be a challenge to deploy that much capital, even for him! Did anyone bother to check if he hit the $50B promised in 2016 before this "double down"? [CNBC]
• At least one survey suggests that many iPhone users don't find the AI features useful or adding much – or any – value. Not all bad news though: they still beat out users of Samsung's Galaxy phones. [MacRumors]
• The third season of The White Lotus hits on February 16 and is looking good. [THR]
• As expected, running out of options, TikTok is now asking the Supreme Court to block the law that will see the service banned. At least until they can hear the case. But it's also risky because if they choose not to block here, it may indicate they're unlikely to overturn in the long run. So it may be up to Donald Trump's "warm spot" to save the service... at least to sell it. [NBC News]
  • Dinner anyone? [CNN]
• "Santa Mode" for ChatGPT is something else – especially if you're just using the service as you normally would and not trying to talk to Santa about holiday stuff. "Santa, explain quantum physics to me..." [TechCrunch]
  • Less fun but more useful are 'Projects' for ChatGPT – essentially a way to create your own folders to organize conversations. Still sort of fun! Kudos to OpenAI for this amazing pace of shipping. [Verge]
• Following up on the 'iPad Fold' rumors, DSCC, the reputable supply chain analysts on the display side, adds more fuel to the fire believing that an 18.8" foldable OLED iPad Pro is on track for 2027... [MacRumors]
• The motions to dismiss the lawsuits related to the bungled sports streaming service Venu were all "DENIED". That keeps it on track for trial in October of next year – which should be timed well for the launch of Disney's new stand-alone streaming ESPN, to perhaps just make this all go away. [Deadline]
• Ghost, the service on which I run Spyglass, now makes it easy to use custom fonts – I've been using my own since day one, but it used to be a bit more complicated, using a customized theme, now it's simple and a nice way to ensure your site doesn't look like every other one on the web. [Ghost]
• Why is Tesla fighting so hard to reinstate Elon Musk's controversial pay package? Taxes, of course. If they just issue a new award to make him "whole", the tax implications could surpass $100B. One way or another, he's clearly getting this money, it's just a question of who will foot that bill – Tesla shareholders, albeit indirectly via dilution, seem most likely right now. [FT 🔒]
• Waymo is going international with their first tests starting in Japan – they're partnering with taxi operators/apps to make it happen. Notably, they drive on the other side of the street, so we'll see what that does to Waymo's brain. [CNBC]
• Per the Netflix and theaters link up top, talks continue to bring Greta Gerwig's take on Chronicles of Narnia to IMAX screens when it opens in December 2026. The only hold up seems to be how Netflix messages this since they don't want to make it seem like they're going back on their streaming-first mentality with other talent. [Bloomberg 🔒]
• Mark Zuckerberg says Threads now has 300M MAU and 100M DAU – the daily number in particular is impressive and they've seemingly been hesitant to share such a number to date. While all of this is of course noted alongside the growth of Bluesky, Threads is seemingly catching up to Xitter faster than expected. [Verge]

I Quote...

"We’ve achieved peak data and there’ll be no more. We have to deal with the data that we have. There’s only one internet."

-- Ilya Sutskever, during his talk at the Conference on Neural Information Processing Systems (NeurIPS). He also noted that, "pre-training as we know it will unquestionably end" – both of which are clear allusions to the "AI Wall" that is currently much talked about (including by him previously).

Sutskever, of course, is somewhat conflicted here given that he was a co-founder of OpenAI who has since left (under reportedly strange strained circumstances) and now has his own new AI startup, Safe Superintelligence, that competes with OpenAI. That startup is probably going to be in a better position if the era of pre-training LLMs is over and they don't have to play catchup from scratch (with a lot of scratch), as xAI, the startup from another OpenAI co-founder you may have heard of has been doing. But that also doesn't mean he's wrong. And, in fact, he's perhaps the best person in the world to speak on this subject, conflict or not.

He had a lot more interesting things to say as well, on agents, on reasoning, on the relationship between human evolution and the brain pointing to where AI may head next. Which is to say, "life, uh, finds a way..."

I Spy...

Ho, ho, ho!