|
The Spyglass Dispatch is a free newsletter sent out daily on weekdays. Feel free to forward it on to others who can sign up and view previous dispatches here. You can also view this dispatch on the web here.
My friend Steven Levy was able to get quite a bit out of the normally beyond-reserved Tim Cook in his latest sit down interview with him. Apple Intelligence. AI fears. Apple's LLM work. The OpenAI would-have-been investment. Vision Pro. AR glasses. Apple's health initiatives. His own time and legacy. Live keynotes! I have some additional thoughts on those answers to each topic...
I Think...💨 He Was Going to Save Intel. He Destroyed $150 Billion of Value Instead. – A few more details in digging into the backstory of what happened with Pat Gelsinger during his tenure as CEO of Intel. First and foremost, to a key point I made a couple days ago around the board: Gelsinger was at first in discussions to join said board, before his vision for the future of Intel impressed them so much that they asked him to take over. And he agreed only after ensuring each board member was fully bought into his plan, which would take years and a lot of capital. In the end, they gave him three years, seemingly in part because he was spending too much capital. Beyond that, there's a bit of history as to why he wasn't set up for success and some bad luck (the Tower Semiconductor deal which China scuttled – the incredible surge of NVIDIA). And he perhaps tried to burn an old, stagnant company a bit too hot with his "torrid" pace. [WSJ 🔒] 💸 The Verge Now Has a Subscription – While I suspect ads will remain the dominant stream of revenue for the site – their content and audience seemingly lines up well for it – this subscription is a very good deal at $7/month or $50/year. Wait – that's less money than I charge! But, as you hopefully know, it's not actually about paying for my content, but rather my time. (Though hopefully the content is worthy too!) The Verge, as a massive site, is obviously quite different and I think making a reasonably priced subscription layer is key. I also sort of backed into it as I was paying for their newsletter bundle, which is now included and was actually more expensive – so I knew it was launching when I got issued a refund for the difference the other day, which is much appreciated! So far I do seem to be seeing fewer ads, which I appreciate even more. [Verge] 🤓 Xreal One Smart Glasses Have Entered the Vision Pro Zone, and I Wore Them — By far the most compelling part of the soon-to-ship visionOS 2.2 update is the new "wide" and "ultrawide" options for using the Mac Virtual Display feature of the Vision Pro. These Xreal One glasses sound sort of like if that was the only feature of the Vision Pro. And it’s… actually fairly compelling. Obviously, they’re far more limited beyond those monitor capabilities, but they’re also far smaller, easier to use, and cheaper at $500 versus $3,500. They also more clearly showcase a future where this is an obvious use-case for such wearables. When they’re not so laborious to put on and use, you’ll just slip on a pair of glasses and get work (or play) done on a massive virtual monitor. [CNET] 🍓 OpenAI is Charging $200 a Month for an Exclusive Version of its o1 ‘Reasoning’ Model – 'Shipmas' is here and the first present under the tree is awfully expensive. It's not clear why this 'ChatGPT Pro' is an order of magnitude more expensive than the 'Plus' subscription as it seems like you'd mainly be upgrading for the 'o1 Pro' model, which is an improvement, but not as much as o1 'Regular' is versus the model that was released as 'Preview' – which, to be clear, seems like an impressive leap. Yes, I know 'Pro' ('ChatGPT Pro', not 'o1 Pro') removes limits on o1 (Regular and mini?) and voice mode too, but those limits seem pretty high given that, as a Pro user myself, I'm not even sure what they are. All of this is nice, but I'm still slightly worried about the increasing complexity of using ChatGPT. Those menus within menus keep expanding... [Verge] 🇫🇷 How Notre-Dame Was Reborn – Absolutely incredible that they pulled this off. And a great read about how they did it. As it gets set to re-open this weekend, even more unbelievable that they did it in the five year time horizon that Emmanuel Macron originally laid out – especially considering that a global pandemic shut down the world for a substantial amount of that time. Everyone pokes fun at the French for perhaps enjoying life a bit too much, but this is a great testament to their will. The best part is about the blacksmiths bonding with the bespoke axes they were creating to cut down the needed trees in just the right way for the recreation of one of the world's wonders. [NYT]
I Note...- It sounds like OpenAI is trying to find a way to alter their prenup with Microsoft to allow them to remain in a relationship even if AGI is achieved. I don't think we should take that to mean AGI is close versus just a gesture to give Microsoft assurances that OpenAI's board isn't going to try to pull a fast one, declaring it achieved, and cutting Microsoft off. Is the romance rekindling? Are Microsoft's own AI efforts not working? Or is it just the reality that OpenAI needs a shit ton of capital and partners are hard to come by at this scale – especially with Elon Musk needing the same capital? As I quipped earlier, these two are the Elizabeth Taylor and Richard Burton of tech. [FT 🔒]
- Intel wasted no time adding two new directors to their board following Pat Gelsinger's ouster. While it sounds like this was more about replacing the semiconductor experience lost when Lip-Bu Tan (now a potential CEO candidate) stepped down a few months ago, the timing cannot be coincidental with the board itself now feeling heat over the fiasco. Former ASML CEO Eric Meurice and Microchip Technology interim CEO Steve Sanghi will join immediately. [CNBC]
- Would the government step in to bail out Intel if things get really dire? Seems highly unlikely, especially with the administration change about to happen. But they could help broker a deal with a preferred partner... [FT 🔒]
- TSMC's Arizona facility is going to help produce NVIDIA's Blackwell chips next year and the plant also apparently counts Apple and AMD as customers. All not great optically for Intel's American production ambitions. [Reuters]
- All of this clearly spells trouble for the CHIPS Act, which the Trump administration might welcome on the PR front, but it will be a nightmare to try to clean up. And if they actually care about bringing production over to the US, tariffs or not, they need Intel here. But there probably isn't even enough time to see what the Trump administration might do if Intel is going to make a big move like selling off the foundry (which their CHIPS grant will make harder to do) and fully outsourcing production (which TSMC probably doesn't even have capacity to handle in full?). What a mess. [FT 🔒]
- Director James Cameron is partnering with Meta to bring his Lightstorm tech to Quest headsets, presumably to battle with Apple's 'Immersive Video' tech for the Vision Pro. No word on any AI angles... [THR]
- Festivitas, for the rest of us – a simple, fun macOS app that lets you add holiday lights to the dock and menu bars. An easy buy (you can pay what you want). 🎄 [MacRumors]
- A 41% jump in viewership for Amazon's 'Black Friday' NFL game this year, plus their rising regular Thursday night game viewership, suggests that people are getting used to watching games on Prime Video. And such "normalization" may actually help Netflix with their Christmas games. [Bloomberg]
- Threads letting users see metrics around individual posts seems like yet another thing that will backfire against Meta in trying to create a vibrant network. Prediction: just like on IG, people are going to delete poorly performing posts and gravitate towards viral, stolen junk (which they're already trying to combat but is everywhere on Threads). Just focus on making it a great network for real time information and conversation! [TechCrunch]
- Bluesky is up to 24 million users, with Jay Graber also giving an update on early subscription efforts and not "enshittifying" the network. [Wired]
- The Joe Rogan Podcast was the number one podcast for the year on Spotify for the fourth straight year. With Call Her Daddy remaining in the number two spot as it seems like the popular shows on the format are getting entrenched. But video remains the wildcard that is rising fast – 1/3rd of the top 50 offered video formats last year versus 1/7th a year ago. That's continued good news for YouTube. [Variety]
- It looks like Humane wants to pivot from their AI Pin to using their 'CosmOS' for other connected devices that want to inject AI. [Verge]
- David Sacks is Trump's pick to be the US crypto and AI "czar" – not sure why we're combining those two very different fields and worlds, but okay. The paid pals continue to rise... [FT 🔒]
- It sounds like Meta's internal coding tool – the cutely named 'Metamate' – uses not just Llama's models, but OpenAI's as well. I think that's a good thing, too many large companies fall victim to the "not built here" nonsense. If you want people to do their best work, let them use the best products and services, not just the best services and products you've built. [Fortune]
- It sounds like Microsoft's Surface Studio line – aka, the big ass touchscreen computers – may be no more. They always looked cool, but it wasn't clear the exact size of the market for such a hybrid? [Windows Central]
I Quote..."[Apple] came to us, and said to us, 'how can you help us with our Generative AI capabilities, we need infrastructure in order to go build,' and they had this vision for building Apple Intelligence."
-- Matt Garman, the CEO of AWS talking about their new Trainium 2 AI chips. This partnership is interesting on a few fronts. First and foremost because Apple rarely lets other companies talk about their work with Apple. And I also suspect they don't love Garman's quote here because it's either the truth or giving Amazon too much credit, implying Apple was in over their heads and couldn't do this on their own. But it's certainly at least in part true since Apple, seemingly, isn't using NVIDIA chips to train their own models – at least not at any sort of meaningful scale. Instead, they were said to be using Google's TPUs. And Apple's own in-house chips, while powerful, clearly aren't built for AI training – at least not yet. And if that's true, it does give some credence to the notion that the partnerships Amazon announced aren't just all for show. One could easily see a world in which Amazon has Anthropic and Apple as their first major Trainium partners. But the real world jury is still out in terms of how the chips will compete against NVIDIA's.
|
