530: Netflix vs Streamers, Anthropic New Models, Is Samsung Slipping?, Tesla's Autopilot, Amazon vs Temu, Film IP + VR, and Stephen King

"I’ve discovered Zuck’s final form"

Error is inevitable. Persisting in it is not.

—Stoic Emperor

☕️🫙👩🏻‍🍳🫖 In the “Well-designed product that makes my life better” category, the Airscape containers by Planetary Design gets my seal (haha) of approval

(btw, this isn’t a sponsored thing, I get nothing, I just like it).

It’s rare these days to find a product that does one thing well and where every detail has clearly been carefully designed to accomplish its mission, I have to highlight this one.

Premise: You’ve got a bunch of loose tea leaves — Japanese Sencha, yum! — or coffee beans — I’m into Ethiopian these days — or maybe even nuts or other staples, and oxygen is bad for them. Over time they oxidize and go stale.

You want to keep them flavorful and fresh for as long as possible.

These containers can help. They’re airtight and have a special plunger lid with a two-way valve that forces air out. You can push it to the level of what’s left in the container, so at first it’s near the top, but it can go all the way down.

It’s very satisfying to hear the “whoosh” of the air going through the valve.

The general principle is sound, but what I like are the details. There are two sizes available. The canister is made of metal and the matte finish feels high quality and looks good (you can get various colors). There’s non-slip rubber on the bottom so it doesn’t move around on your kitchen counter. The non-plunger lid has a rubber lid for extra airtightness.

These things are a little expensive, but they should last forever. If they help make my coffee and tea better over decades and allow me to buy larger quantities (ie. at lower unit cost) because I know it won’t go stale, it feels like a good investment.

🕺 I think I’ve discovered Zuck’s final form in 10 years:

🛀💭🤖🎨🖼️ I’m a little surprised by the lack of attention given to generative AI models for images compared to text models. They still get some attention, but somehow it seems like we’ve all gotten used to them and the focus has mostly shifted to other things.

I would not have predicted this 2 years ago if you had shown me an image from today’s models versus what was state-of-the-art at the time (remember how freaky and strange those early images could be?). I would have been amazed and predicted it would make big waves in the zeitgeist since images are more popular online than text.

---

🏦 💰 Liberty Capital 💳 💴

📺 Streaming Platforms Paid Subs + Netflix Case Study 📈📉

I’m guessing there is no Prime Video and YouTube because they aren’t “paid platforms” like the others.

Prime Video is bundled with shipping and other perks. How many subs are active? How man wouldn't subscribe to the standalone video product?

YouTube is mostly a free platform and mostly user-generated (sometimes I wonder how much of YouTube is someone watching someone else's YouTube content and reacting/commenting?), so while it’s huge and *does* compete with the others, it’s not a clean comparison.

Also, as some have pointed out on Twitter, the chart above shows why investing is hard — that little dip in Netflix subs around 2021 led to this reaction for the stock:

That’s a 76% drawdown! 😱

It was apparently too much to take for many:

Bill Ackman sold his entire stake in Netflix on April 20, 2022, just one day after Netflix reported its first-quarter earnings and disclosed a significant loss of subscribers.

Source.

🇰🇷 What is going on at Samsung? Is Sam Slipping? 🐜💸

For many years, Samsung seemed like the Chosen One.

Or at least, the Chosen Second One…

They were a strong #2 to Apple in phones, a strong #2 to TSMC when it came to fabbing, and a strong #1 or #2 in various other parts of the supply chain — memory, display panels, smartwatches, NAND, etc.

But in recent times, they’ve been slipping in multiple ways…

They kind of missed the AI wave, which creates a vicious cycle for them. Competitors who are riding it — like TSMC on the foundry side and SK Hynix and Micron on the HBM RAM for data-center GPUs side, for example — have extra revenue that they can reinvest into capex and R&D, helping them pull further ahead.

Samsung shouldn’t be counted out. They are still a giant with the resources to catch up. But there are signs that something is going on and morale on the inside may be bad:

[For[ a recent SK Hynix job posting looking for three experienced etching engineers, around 200 engineers currently employed at Samsung applied.

This means that most of Samsung’s engineers working in its fab lines that qualified for the job applied. They also said the information was shared widely in the industry as it was highly unusual for so many Samsung applicants to do so.

Samsung engineers with relatively little experience are also transferring more to SK Hynix, which runs a program that recruits chip engineers with less than five years of experience. [...]

the Korea Electronics Technology Institute recently made a job posting looking for three researchers. The posting attracted around 50 doctoral-level engineers from Samsung’s chip business, they also said. Additionally, eight recently hired employees by KETI were all from Samsung as well,

This wouldn’t be as notable in Silicon Valley, but in South Korea, moving around from one company to another isn’t as common.

Last month, in a town hall meeting between executives and employees, a Samsung employee asked whether the company had a solution for securing and retaining talents. Samsung memory boss Lee Jung-bae told them to work hard and asked them to encourage those thinking of leaving to stay.

Samsung’s annual salary to employees in the past was the industry’s highest but today only with bonuses are their salary a little better than their competitor’s

Compensation may be another factor, but that’s another area where riding the AI wave with those sweet margins can help be more competitive…

👀🤖🚘 Tesla’s ‘Autopilot’ Safety Statistics 💥😵

Let’s have a look at Tesla’s latest safety report for self-driving:

In the 3rd quarter, we recorded one crash for every 7.08 million miles driven in which drivers were using Autopilot technology. For [Tesla] drivers who were not using Autopilot technology, we recorded one crash for every 1.29 million miles driven. By comparison, the most recent data available from NHTSA and FHWA (from 2022) shows that in the United States there was an automobile crash approximately every 670,000 miles.

At first glance, this is very impressive.

But let’s put on our Bayes’ #1 Fan hat for a sec 🧢🧐

Is this 🍎 vs 🍎?

Drivers decide when to engage self-driving mode. They are very likely to do it when driving conditions are better and in areas that are easier to navigate. They are also likely to stay fully engaged and drive themselves when road conditions are worse and in more difficult areas.

So comparing accidents when self-driving is engaged to the general population’s driving, which includes both good and bad conditions and areas, isn’t quite fair.

We could also question whether people who buy Teslas are a self-selected group of drivers who are less prone to accidents generally: higher income + the fleet is fairly recent since high volumes of Model 3 and Y only truly hit the market in the past 5 years, so there are few old clunkers which are inherently less safe.

That doesn’t mean that self-driving isn’t mostly safer than most human drivers, just that these statistics need to be looked at in context.

🛒🇨🇳 Amazon’s Temu Competitor 😯

Is that a $20 sofa?

I was reading about Amazon’s counter-punch for Temu (or their way of making the government close the loophole that allows Temu to ship small packages into the US without much friction from customs…) and almost did a spit-take when I got to this part:

As Amazon prepares to launch a new low-price storefront to combat Temu, it’s imposing severe price caps on what merchants can charge for their wares on the outlet, including an $8 limit for jewelry, $9 for bedding sets, $13 for guitars and $20 for sofas

$13 guitars?!

$20 sofas!?

What’s going on here?

As my friend MBI (🇧🇩🇺🇸) joked, maybe they’re getting inventory by buying used crap on Marketplace…

---

🧪🔬 Liberty Labs 🧬 🔭

🥽 🎬🎞️🎥🍿 Film IP + Virtual Reality 🤔

I was thinking about VR. What would I want to “visit”?

One obvious thing is the locations shown in various films. If I could, I would love to walk around the 1955 town square in Back to the Future, or the top of the Cliffs of Insanity where the sword fight takes place in Princess Bride, or to watch some dinosaurs walking around Jurassic Park, or walk around the Millennium Falcon, or the Arrakeen royal Palace in Villeneuve’s Dune, or the cyberpunk LA of Blade Runner 2049, with purple Joi holograms talking to me.

Or any set from any Wes Anderson film! Imagine getting up close to all the details…

TV could also offer a rich vein. How about the house in Six Feet Under? Or Al’s saloon in Deadwood 🤠 Or the Serenity from Firefly?

I wouldn’t even need to *do* much. I just want to be there. No need to gamify or storify. I’d just like to visit these places that I feel familiar with because of the films.

Thanks to LLMs, we could populate these environments with NPC characters that you can interact with, and because they are trained on the script and the film, they could probably do a decent job.

Owning that film IP for VR could be valuable because it’s scarce (people want to be on the bridge of the USS Enterprise, not a generic spaceship), and because there’s a social element to it (being able to tell your friends and they know what you’re talking about because they’ve seen the media on which the experience is based).

Building all this manually would be a tremendous amount of work, but I can imagine that AI will get good enough that someday we’ll be able to feed a film into it, along with all the development assets owned by the studio that made the film (concept art, 3D digital assets used to make environments, floorplans used to build sets, the script, etc) and maybe a large portion of creating virtual environment could be automated 🤔

Wouldn’t that be cool? I’d definitely buy a VR headset to experience that!

🤖 Anthropic Upgrades Sonnet 3.5 + Introduces “Computer Skills” for Models 💻

Oh wow, this is big!

Claude 3.5 Sonnet is still my favorite model — it has the “flavor” that I like best — and it just got better! Significantly better, based on the numbers above, but we’ll have to see how that translates in practice.

(I wish they had changed the version number — this should be Sonnet 3.6 so that we know for sure which version we’re using)

The only big contender missing in the comparison is OpenAI’s o1 and o1-Mini, but these 🍓 models work differently enough that it’s probably fair, even if I still wish they were included.

Anthropic is *crushing it*.

GPT-4o is starting to fall further behind and I wouldn’t be surprised if this put pressure on OpenAI to release a more competitive model soon (GPT-5 is probably not ready, but they may have some kind of GPT-4.5o in the oven…

Anthropic also released a smaller model called Haiku 3.5, and it appears to be competitive too, though Gemini 1.5 Flash does better across the board. It’s not clear if Flash is a bigger and more resource-intensive model, though.

The next 🤯 thing is that Claude can now use your computer autonomously (well, kind of):

Instead of making specific tools to help Claude complete individual tasks, we're teaching it general computer skills—allowing it to use a wide range of standard tools and software programs designed for people.

To make these general skills possible, we've built an API that allows Claude to perceive and interact with computer interfaces.

It’s still just an early step and don’t expect to do too much with it yet, but it’s very impressive! Ethan Mollick has tried it and has more details.

---

🎨 🎭 Liberty Studio 👩‍🎨 🎥

👀🐲 ‘The Eyes of the Dragon’ by Stephen King (not horror) 📗🛡️⚔️

I’ve been reading this book to my 10-year-old for a few weeks.

It’s epic fantasy in a quasi-medieval setting, and a departure for King, as it’s not horror and targeted to a younger reader.

At first, I thought I had made a mistake.

Why?

In my memory, this was a short & easy read. However, it turns out it’s longer and more detailed than I remembered. I was afraid it may be a bit too much for my boy, but he loves it and whenever it’s time to stop reading, he usually asks for a few more pages (always a good sign!).

It’s showing me that he’s ready for more ‘advanced’ books, which is great!

Next on my list is ‘The Princess Bride’ by William Goldman. We both love the film and can quote it at each other, so this is a no-brainer. The book is similar to the film, but as books do, it adds a lot to it. And Goldman is such a fun writer.

Also on my list is another horror writer who wrote a kids’ book: Clive Barker’s ‘The Thief of Always’. I think that one is shorter and simpler than ‘Eyes of the Dragon’, but I may also be misremembering it.

If you’re looking for something slightly off the beaten path for your kids, this may be a good choice.

Trivia about the link between this book and Misery:

Some of King's established fans rejected the novel, considering it a children's book – King originally wrote the novel for his children. Another reason for fan rejection of The Eyes of the Dragon was the fact that it was epic fantasy, with little to no elements of the horror that typified King's most successful work of this era. 

Negative fan reaction to The Eyes of the Dragon was an inspiration for King's subsequent novel Misery.