378: Stripe + Amazon, RxPass, Microsoft + OpenAI, Texas Instruments, Toyota, Saunas, StableDiffusion, and John Carmack
"if you never disagree with anything..."
You can’t know *everything*, but you should convince yourself that you can know *anything*. Any one particular thing, if it was created or discovered by humans, you can find out what you need.
You can know it deeply. You can drive a nail through whatever layer-cake problem space you’ve got, and learn a cross-section. –John Carmack
🧙♂️🧛♀️🧟♀️🔮🎲🥃🔥 My weekend away with 5 friends was a lot of fun. We’re thinking of making it a recurring tradition.
We rented a great log cabin at Fiddler’s Lake and spent a couple of days making jokes, eating lots of great food *and* junk food, playing D&D, and relaxing in the hot tub and sauna.
The science appears fairly strong that frequent sauna use has significant health benefits, so I’ve been meaning to experiment with saunas for a few years, but never really got around to it (the pandemic didn’t help). Rhonda Patrick has compiled a lot of information on this page, and there’s a video version, if you’re curious.
This weekend was the perfect opportunity!
I went multiple times, including once where I did over 30 mins at 80-85 Celsius/176-185 Fahrenheit. It may not sound like much to you if you’re a pro, but for a newbie, it was hot!
As someone who’s very sensitive to hot & cold, I wasn’t sure if I could handle it, but it was ok as long as the air was dry. When we experimented with adding humidity, things got much harder — it felt like being punched in the face by the heat!
I suspect that I’ll have to figure out the proper ratio of heat + humidity over time, as my body becomes more heat-adapted, because adding humidity when the air was already 180°F was just too much for me at this point.
This experience has made me bump up the “home-sauna project” on the “future plans” list. I’d love to have a barrel sauna in my backyard, although these things are rather expensive (I can’t help but think “why does it cost 1/3 of a car for a pile of cedar planks and an electric heater?”).
I wonder how difficult it would be to build one from online plans & instructions for someone who knows how to work with wood (I’m not talking about myself, but I know some people). I’m sure you can buy the electric heater separately. I’ll have to look into that 🤔
The huge pepper mill (32 inches tall!) in the bottom-left photo was a gag prize that my friend got for the group. I ended up winning it! For years, I’ve been talking about how much I like having a nice pepper mill, but this monster is something else 😬
The same very generous friend also brought a bottle of Ardbeg Uigeadail to share. It’s in my top 3 fave scotches, possibly #1 (I can’t decide!) 🥃 🔥🥇
🤖📸🎭 Here’s a generative AI use case that randomly popped into my brain (it may already exist, I don’t know): Anonymizing public photos/videos by slightly altering faces.
Let’s say I want to post a photo of myself publicly, and I want to show my face, but I don’t want people to be able to use facial recognition to track me down, or even be 100% sure that it’s me.
In other words, the point is to show the whole scene, not necessarily to associate it with me. My identity isn’t important to the photo, but it would look worse if it was cropped or edited in a way that would remove or obscure my face.
Maybe I could use some type of generative AI software that would slightly alter the features of my face so that it looks somewhat familiar, but different enough to fool people and facial recognition software.
✍️🤖 Just got on the beta for Notion’s AI assist tool.
I kinda feel like *all* productivity tools will have a menu like this soon. It’ll be table stakes (which means a lot more demand for compute in the cloud — somewhere Jensen is smiling):
As I wrote about in edition #376, GPT’s natural home is not a search engine, it’s Microsoft Office (aka Clippy that doesn’t suck 👀📎).
🛀 It’s hard not to get used to things over time, not to anchor and re-anchor on whatever is familiar…
For example, I try not to merely glance over the employee numbers at big companies, but rather, really think about how many people we are talking about.
Any way you slice it, 30,000 is a lot of people!
Many of humanity’s greatest accomplishments have been achieved by small groups of people.
It doesn’t mean that what these large organizations are doing is the same kind of work and could be done by small groups, but I try to remember that a lot can be done by few people if they are highly effective, and conversely, it’s easy to let friction and bureaucracy turn large groups into ineffective entities.
💚 🥃 🐇 This is the free edition of Liberty’s Highlights with 14,400+ subscribers.
You can get 2 extra editions/week full of juicy stuff + access to the private Discord 🗣🗣🗣 community by becoming a paid supporter (it’s quick & easy).
Paid post since last week 🐇:
Many cool things in this one, but the one I’ll highlight now is the video that shows the clever tech that makes it possible for helicopters to land on ships in choppy seas. Really cool stuff. 🌊🚁🌊
If you click the links above, you can see the intros for free and there’s a link to get a 7-day free trial of the paid version. What have you got to lose?
🧩 A Word From Our Sponsor: Heyday 🧠
Do you have 100+ browser tabs open right now? 😬
Give your memory a boost with Heyday, the research tool that automatically saves the content you view, and resurfaces it within your existing workflows. 👩💻
It’s like cheat codes for your memory. 😲💡
🏦 💰 Liberty Capital 💳 💴
🛒 Stripe deepens partnership with Amazon 💳 🤔
Under the new agreement, Stripe will become a strategic payments partner for Amazon in the US, Europe, and Canada, processing a significant portion of Amazon’s total payments volume across its businesses, including Prime, Audible, Kindle, Amazon Pay, Buy With Prime, and more.
Stripe will expand its use of Amazon Web Services, Stripe’s long-standing cloud infrastructure provider, to run and grow its business while reliably serving millions of internet companies.
I’m guessing that this is a better deal for Amazon than Stripe, since there aren't that many companies with Amazon's volume for Stripe to win, but Amazon could keep using other processors as it has. It’s no doubt a race to the bottom among payment processors to win this business.
Amazon has the leverage in this relationship.
So it’s probably low margin for Stripe, maybe break-even. Possibly a defensive move just to keep that volume away from competitors..?
What does “a significant portion of Amazon’s total payments volume” really mean? Is 5% significant? 10%? 20%? More? Less? ¯\_(ツ)_/¯
Meanwhile, Amazon is probably getting a great deal on payment processing and will make money on the AWS side too.
I have to wonder if this could be a sign that Stripe’s relationship with Shopify is about to change — as Stripe’s biggest customer, Shopify may have had leverage to try to keep Stripe from chasing Amazon in exchange for exclusivity — maybe something is going on behind the scenes and just hasn’t been announced yet 🤔
💊 ‘Amazon adds RxPass generic prescription perk for Prime members’ 📦
Speaking of Amazon, they just launched a $5/month add-on to Prime:
RxPass, a new Prime membership benefit from Amazon Pharmacy that offers patients affordable access to commonly prescribed generic medications that treat more than 80 common health conditions.
With RxPass, Prime members can receive all of their eligible medications for one flat, low monthly fee of $5, and have them delivered free of charge. There are no hidden fees and no markups to the $5 per month subscription.
RxPass is available starting today in most U.S. states.
This covers about 50 generic medications.
I’m not sure how good a deal this is, as I’m not familiar with the alternatives offered in the US, but for those who subscribe, it will be one more incentive to not churn from the Prime bundle.
🤖💰💰💰💰💰 It’s official: Microsoft makes huge investment in OpenAI
We’ve known about it for a couple of weeks, but it wasn’t official until now (I wrote my thoughts about the strange deal in edition #374):
we are announcing the third phase of our long-term partnership with OpenAI through a multiyear, multibillion dollar investment to accelerate AI breakthroughs to ensure these benefits are broadly shared with the world.
This agreement follows our previous investments in 2019 and 2021 [...]
Supercomputing at scale – Microsoft will increase our investments in the development and deployment of specialized supercomputing systems to accelerate OpenAI’s groundbreaking independent AI research.
New AI-powered experiences – Microsoft will deploy OpenAI’s models across our consumer and enterprise products and introduce new categories of digital experiences built on OpenAI’s technology. This includes Microsoft’s Azure OpenAI Service
Exclusive cloud provider – As OpenAI’s exclusive cloud provider, Azure will power all OpenAI workloads across research, products and API services.
They didn’t give details on the financial structure of the deal or even the dollar amount, only that it’s “multibillion” — we can guess that it’s similar to what was previously reported ($10bn over some years).
Back in November, Microsoft and Nvidia announced a new super-computer using “tens of thousands of H100”. This thing is really going to scream, especially considering that H100s are particularly good at training Transformer models (between 1.5 and 6x faster than A100), which is what is used by LLMs like GPT-4/ChatGPT.
I wouldn’t be surprised if OpenAI is major a beneficiary of this big iron.
Altman is quoted in the release: “Microsoft shares our values and we are excited to continue our independent research and work toward creating advanced AI that benefits everyone.”
I don’t think the word “independent” was thrown in there casually.
This must be part of the deal — they get a ton of compute credits and some money to hire talent, and Microsoft gets a bunch of the upside of commercialization of whatever AI tools/products they come up with, but they must’ve included in the deal that they keep operating independently from Microsoft and can steer the research (especially into AGI, which is the whole point of OpenAI’s existence).
But who really knows ¯\_(ツ)_/¯
‘Microsoft has laid off entire teams behind Virtual, Mixed Reality, and HoloLens’ 🥽✂️
Despite having acquired AltSpaceVR back in 2017, Microsoft culled the entire team behind the virtual reality workspace project this past week. As a result, AltSpaceVR will shutter for good in March, effectively ending Microsoft's "metaverse" efforts with it. [...]
In addition to the death of AltSpaceVR, Microsoft has also culled the entire team behind the popular MRTK framework. MRTK is Microsoft's "Mixed Reality Tool Kit," which is a cross-platform framework for spatial anchors in virtual reality spaces. MRTK was built for Unity VR integrations, and works with Meta's headsets with a focus on HoloLens.
HoloLens has been scaled back already in recent years following the departure of its chief architect Alex Kipman. Microsoft has been pursuing a HoloLens contract with the U.S. military, which was recently scaled back by the U.S. Congress, owing to reported problems with the program. (Source)
This could mean that Microsoft prefers to let
It’s probably a better business anyway and plays to Microsoft’s strengths.
🐜 Texas Instrument CEO Rich Templeton to become Chairman, COO Haviv Ilan to become CEO
Templeton is a great CEO, and after 19 years with the top job, he’s kicking himself up to chairman — which he had attempted to do before, but he had to come back when his successor left after 2 months because of violations related to "personal behavior".
His successor as CEO will be the company’s COO, which is similar to how Templeton got the job after being COO for 4 years back in the early 2000s.
Ilan has been with TI for 24 years, so the company is very much continuing the tradition of internally developing leadership and having people who know the business intimately and have experienced many analog semiconductor cycles at the helm.
A Toyota Prius… that looks good?! (but no electric-only version) 🔥
I can't believe it.
Toyota has made the new Prius actually look good!
If only they had made a fully electric version of it…
They’ve really squandered the lead that they had in electrification, being an early player in hybrids, having more experience with electric drivetrains and batteries than other automakers. But they never really did much with it when it comes to BEVs, and now they’re behind.
There’s a plug-in hybrid version is coming with 50% more electric range than the previous one, but that’s still only about 60 km/37 miles.
🧪🔬 Liberty Labs 🧬 🔭
🏞️🎨🤖 StableDiffusion for Dummies: How AI Image Generation Works 🧙♂️🪄
It looks like magic, but it isn’t…
Not *entirely*, anyway.
🍎 Apple M2 Improvements over M1 🔥
Ben Bajarin poked around the new M2 Max and found some improvements compared to the previous generation.
Here are the highlights:
More effecient P-cores in M2 next-gen CPU
More performant E-cores in M2 next-gen CPU
Better thermal management overall in M2 Pro/Max
Better load balancing across all cores
We probably won’t see a jump forward like we saw going from x86 to the M1 for many years, but it’s nice to see that Apple’s silicon team keeps relentlessly pushing to squeeze as much performance and efficiency out of these time slices of polished sand.
Interview: John Carmack (Podcast & Video)
I enjoyed this interview with John Carmack, an engineer’s engineer and the co-creator of the games that shaped my teen years (Doom & Quake).
I disagree with some stuff that is said, but hey, if you never disagree with anything, either you’re not thinking independently, or you don’t expose yourself to enough viewpoints!
He discusses his history with technology, going through how he started to code, some of his classic technical hacks to create groundbreaking games (the Keen games, Wolfenstein 3D, Doom, Quake..). He also talks about his VR works at Oculus/Meta, his current work in artificial intelligence, and his other big interest in nuclear fission (how to make it cheaper).
The discussion about AI near the end of the podcast is quite interesting, but it’s weird how it feels like it’s older than 5 months (the podcast was released in August 2022) because so much happened since — Midjourney, StableDiffusion, ChatGPT.
🎨 🎭 Liberty Studio 👩🎨 🎥
🎥🎬🎤🙊 Explainer on modern TV/Film audio — Why dialogue can be hard to hear, mixing, etc
I thought this was an excellent behind-the-scenes look at modern audio recording and mixing, and what compromises are made between clarity and better performances.