504: Deep Dive in Apple's AI Strategy, OpenAI, Nvidia's Jensen and Patrick Collison, Elon Musk, Google TPU, KIA EV3, and Paperman

"it’s like a glass that refills itself."

Haters are a good problem to have. Nobody hates the good ones. They hate the great ones. 

—Kobe Bryant

♟️ When people think someone is playing 4D chess, they usually aren’t.

When someone *is* playing 4D chess, it usually goes unnoticed.

📬 If you’re a paid supporter of this steamboat, you should’ve received an email from me last week telling you about our first Zoom Q&A/chat/hang out session (June 12 at noon ET, for an hour — TODAY!).

If you didn’t, check your spam folder or contact me and I’ll send you the call info.

🛀💭🫀❤️‍🩹❤️‍🔥 How much love is in your heart?

The more you love you spend, the more love will be in your account.

Love is not a consumable, it’s like a glass that refills itself.

The faster you drink, the faster it replenishes.

(I know, it sounds cheesy, but whatever, it’s true)

💚 🥃 🚢⚓️ If you’re getting value from this newsletter, it would mean the world to me if you become a supporter to help me to keep writing it.

You’ll get access to paid editions and the private Discord. We’re also starting to do Supporter Zoom calls.

If you only get one good idea per year, it’ll more than pay for itself. If it makes you curious about new things, that’s priceless.

---

A Word from our Sponsor: 💰 StockOpine 💰

→Struggling to Pick Quality Stocks for the Long Term?←

At StockOpine, we do the heavy lifting so you can save time and make informed decisions. Our team dedicates countless hours to analyzing and selecting top-performing companies, providing you with actionable investment ideas.

Why Choose StockOpine Premium? By joining, you’ll gain access to:

• Two In-Depth Company Analysis per Month. 📑📑🕵️‍♂️

• Quarterly Portfolio Updates + Short Investment Thesis for New Positions. 💡👍

• Earnings Reviews + Industry Overviews. 📊📈🏦

• Monthly Valuation Snapshots. 🗞️🚨🗓️

⭐️ Sign up now and get 25% off FOR LIFE with the coupon code “LIBERTY25” (good through June 30th, 2024). ⭐️

---

🏦 💰 Liberty Capital 💳 💴

🗣️🗣️ Nvidia CEO Jensen Huang Chats with Stripe CEO Patrick Collison 💳 🐜

I could probably leave it at that. It’s worth watching just to see these two interact.

The section where Patrick asks about Nvidia’s relatively flat-org structure, and how Jensen has 60 direct reports which wouldn’t generally be considered “best practices” is particularly thought-provoking.

Jensen's approach may not work for everyone, but his defense of it is great because he has clearly thought it through and created a system that works well for him and his company. He didn’t just cut & paste what others are doing or what would be considered kosher in a Harvard Business School case study.

Jensen also mentioned that he was also a fan of Perplexity and “uses it all the time, even when I know the answer, I want to see what it will come up with.” Right on 🤘

h/t friend-of-the-show Nick Ellis (🎩)

🍎🤖📲💻 Apple sprinkles AI everywhere at WWDC: High-Level Strategy Highlights 🔍

Back in Edition #487, I speculated about Apple’s potential AI strategy.

To paraphrase myself a bit, I said:

• If I were Apple, I’d create a kind of “air traffic control” UX layer that would abstract away other people’s models and focus on the *user experience* of using them. My model would decide whether to use smaller on-device models or bigger models in the cloud, and make everything modular so that a constellation of specialist models could be used for certain tasks and multiple foundational models could be used. This would keep them competing for the workloads and I’d always have the best one available whoever is in the lead. I’d specialize on the top level user experience, which can be differentiated.

Well, it looks like that’s pretty much what Apple has announced at its Worldwide Developer Conference (WWDC) on Monday. 🎯

They branded it all as “Apple Intelligence”, which isn’t a great name, but whatever.

They have smaller specialist on-device models that can do all kinds of text and image generation and manipulation (summarizing text, rewriting text, generating images and emojis, editing photos, etc), and an updated Siri that can access app APIs to do actions on the user’s behalf.

Because of their low-level OS access, they can be very agentic. Only Google and Microsoft can even hope to compare, but their ecosystems aren’t as integrated.

Andrej Karpathy summarizes it well:

Step 1 Multimodal I/O. Enable text/audio/image/video capability, both read and write. These are the native human APIs, so to speak.

Step 2 Agentic. Allow all parts of the OS and apps to inter-operate via "function calling"; kernel process LLM that can schedule and coordinate work across them given user queries.

Step 3 Frictionless. Fully integrate these features in a highly frictionless, fast, "always on", and contextual way. No going around copy pasting information, prompt engineering, or etc. Adapt the UI accordingly.

Step 4 Initiative. Don't perform a task given a prompt, anticipate the prompt, suggest, initiate.

Step 5 Delegation hierarchy. Move as much intelligence as you can on device (Apple Silicon very helpful and well-suited), but allow optional dispatch of work to cloud.

Step 6 Modularity. Allow the OS to access and support an entire and growing ecosystem of LLMs (e.g. ChatGPT announcement).

Step 7 Privacy. <3

They spent a lot of time on privacy, claiming that while the AI models have access to personal data for context, it remains on-device for most tasks. If it needs to go to the cloud because a larger model is required, it’ll run in Apple’s cloud on Apple Silicon chips (this has been reported as being M2-family chips — I’m sure they’ll create custom chips for that over time) that have all of Apple’s security features like the secure enclave, and that the data won’t be stored.

They said that their code for the cloud will be audited and that the workloads will only run on servers that can cryptographically prove they’re running that code.

It sounds good, but I’ll be curious to see these audits and hear more from security experts about the actual implementation once iOS 18 is out.

🍎🤝🤖 OpenAI Partnership… the first of many?

The other big announcement was the partnership with OpenAI.

It was only revealed after spending a lot of time on Apple’s own models and features. Clearly, they want to put the spotlight on their own AI models (the primary local one is pretty small at 3bn parameters, closer to Microsoft’s Phi). More details from Apple here.

To Apple, OpenAI is just one of many potential suppliers rather than a close symbiotic partner like for Microsoft+OpenAI (though Microsoft is trying to hedge its bets lately).

The way they framed it was brilliant (for them).

They said: “We want you to be able to use these powerful external models without having to jump between tools, so we’re integrating them right in your experiences. And we’re starting with the best of these, ChatGPT from OpenAI, powered by GPT-4o.”

Whenever ChatGPT is invoked, you get a notification and decide if you want to send the query or not. It is NOT transparently integrated (more on that below).

ChatGPT will be integrated into the system-wide AI tools.

ChatGPT and DALL-E will be available for free in Siri and system-wide in writing tools. Paid users will be able to log in with their accounts to access paid features through Apple’s various services.

Apple said that even for free users, their “requests and information will not be logged”, which was no doubt one of the things they negotiated with OpenAI because free ChatGPT accounts don’t usually get that.

Apple also mentioned that “we also intend to add support for other AI models in the future”, which I took to mean that it’s likely that they will also offer Gemini if they can figure out a deal with Google. But this could also mean Anthropic and Mistral and Llama… 🦙

Maybe all of them over time, fully commoditizing these suppliers and switching between them like how a large multi-national retailer may switch between payment processors and send a varying ratio of transactions to each depending on various factors (ie. 30% to GlobalPay and 40% to Adyen, etc) .

Final thoughts on Apple AI 🤔💭

I think it’s a strong strategy. Top AI labs will keep pushing the bleeding edge of large models forward, and Apple can reap the benefits from all those billions of R&D by integrating these models into its higher-level AI front-end UX.

I’m sure Apple will also develop a lot of its own models over time. They may not be on the bleeding edge, but they can still be very useful and finely tuned to its users’ needs. F.ex. I’m sure the Siri intent model will get better over time at doing what you ask it to.

A lot of the features that they announced won’t blow people’s minds in demos, but they are things that billions of people do every day, and that tends to matter more over time.

They sprinkled AI magic in iMessage, in Mail, in the Photos app, in Notes, in Calendar, and that’s where most people spend most of their time. And a useful Siri that understands what you mean and does it, and answers your question properly — that’ll be so useful and bring the AI Chatbot experience to hundreds of millions of people who aren’t otherwise using ChatGPT or Gemini.

Apple is positioning itself as an AI aggregator, and I wouldn’t be surprised if OpenAI doesn’t even get paid by Apple.

I would guess that the exchange of value is that Apple gets a powerful model to supplement their own smaller specialist models, and OpenAI gets a huge top of funnel for customer acquisition and will try to convert as many free users into paid ones as possible (ie. once GPT-5 is out for paid users only, they may have negotiated an upsell flow with Apple where the ability to upgrade will be offered).

‘Elon Musk threatens to ban Apple devices from his companies over OpenAI partnership’ 🍎✋

Musk, who clearly hasn’t watched Apple’s presentation (but I get it, he’s busy), tweeted this:

If Apple integrates OpenAI at the OS level, then Apple devices will be banned at my companies. That is an unacceptable security violation.

And visitors will have to check their Apple devices at the door, where they will be stored in a Faraday cage

It’s patently absurd that Apple isn’t smart enough to make their own AI, yet is somehow capable of ensuring that OpenAI will protect your security & privacy! Apple has no clue what’s actually going on once they hand your data over to OpenAI. They’re selling you down the river.

Additional context for this is that Musk is in a feud with OpenAI and competing with them for AI talent, and he’s been competing with Apple in the past when they were working on an EV. So he’s not exactly a neutral observer…

In any case, Apple DIDN’T integrate OpenAI at the OS level.

They’re not giving OpenAI access to your phone’s data, they’re not integrating OpenAI models in their apps (unlike Microsoft — can Musk’s employees use Microsoft software? Copilot?). All the AI features announced are run on Apple’s own models, most of them on-device, and a few in Apple’s own cloud.

And then sometimes you can invoke ChatGPT as a module to do things like answer questions in Siri or compose text in an email or whatever. Each time you have to click on a button to decide to send something to ChatGPT and only that specific info is available to the model. Apple claims that OpenAI will not log those queries, a condition that was no doubt part of their deal with OpenAI (that remains to be verified).

🤖🫳🐜 OpenAI is Poaching from Google’s TPU Semiconductor Team 🤔

Friend-of-the-show Dylan Patel writes (🐜🔍🕵️‍♂️):

Recently, OpenAI has gone on a poaching spree growing their team from just a few to double digits [...]

OpenAI is pillaging the Google TPU team of some of their best talent. Almost everyone who has been poached either works on the Google TPU today or has in the past.

Google’s TPU team is a prime target because Nvidia is a partner of OpenAI and the last thing they want is to piss off Jensen, while Google is a direct competitor.

Part of why they’re able to attract such quality talent is because they’re working on the largest compute systems in the history of humanity, and such ambitions are very powerful in attracting talent that can decide where to work and will be well-paid regardless, so they end up favoring the most interesting problems to work on (as we’ve seen with Elon Musk’s companies).

The quality of people they are picking up is amazing. When we ask their peers about these people, often the response is in the realm of “they are one of the best engineers I’ve ever worked with.”

OpenAI also doesn’t want to rely too much on Microsoft’s own custom silicon, as that would further reduce their leverage in their relationship, and Microsoft is already spreading its bets around and likely won’t make silicon that is fully optimized for just OpenAI’s models.

In any case, as a silicon enthusiast, I’m looking forward to what they come up with.

🦅🇺🇸 U.S. GDP Growth vs Rest of G7

the US has seen real GDP grow by nearly 9% since the pandemic began, by far the strongest of any of its G7 peers, which have averaged only 2.7%. Canada’s economy has been the next best in the group of seven, growing 5%.

Longer-term trends aren’t great for Europe either.

As I was growing up, I constantly heard about the decline of the United States.

And to think, all this time, it's Europe I should have been worried about.

Don’t get me wrong, there’s plenty wrong with the U.S., but there are many pockets of dynamism and it still attracts a lot of the world’s talent. I’m not seeing as much dynamism from Europe, which is sad.

Stagnation tends to lead to zero-sum games, and that rarely ends well.

Source.

---

🧪🔬 Liberty Labs 🧬 🔭

🔌🚘🔋 EV3: Preview of KIA’s new Electric Car 🇰🇷

I love my EV6.

It’s a great car, and I don’t see myself ever going back to internal combustion.

I recently saw that a refreshed version of it had been unveiled — I don’t think the aesthetic changes are an improvement over the old look, but I wish I had the new bigger battery (84 kWh replacing the previous 77.4 kWh).

KIA and its sister company, Hyundai, are doing a great job with EVs and hybrids these days. The latest model to be unveiled is the smaller EV3. It looks more like a smaller EV9 than a smaller EV6, which isn’t a bad idea since the boxier proportions will maximize interior space.

🔓 The most common 4-digit PINs visualized 🔑

“According to the analysis, just 20 4-digit numbers account for 27% of all PINs”

If your PIN is among the “most common” listed, consider changing it to one of the higher-entropy “least common” ones. Or rather, consider getting a PIN longer than 4-digits, because that’s really not secure.

Now that phones have biometrics and we don’t have to enter PINs very often, you really should use a longer PIN (8-digits) or even a passphrase.

Source.

---

🎨 🎭 Liberty Studio 👩‍🎨 🎥

💋📄 ‘Paperman’ — A Great Animated Short 🎞️

It’s very short, no excuse. Check it out!