351: Amazon Belt-Tightening, FTX, TSMC, Twitter, Google vs JPEG XL, Vitamin D, and Midjourney
"have fun, learn, explore and create"
The most dangerous error is failure to recognize our own tendency to error. That failure is a common affliction of authority. —B.H. Liddell Hart
🐦 🤕 Any move Twitter makes will be instantly battle-tested by millions of trolls. That’s why the easy-to-get verified checkmarks were *immediately* widely abused.
If you’re a startup and make a mistake, because you have few users and they mostly tend to be your ideal customers who love you enough to be early-adopters, they’re a lot more forgiving.
Twitter’s hordes, not so much.
mistake x scale = impact
While iterating fast can be great, doing it on a mature platform presents large challenges. It’s why we so rarely see it happen.
I really hope Musk and his team (he has a team, right?) figure it out quickly, because Twitter is special, and there’s nothing waiting in the wings to replace it.
🐜🐘 Once in a while, I catch myself thinking small.
Like, wow, we’re reaching 13,000 subscribers, that’s a lot!
Pff, that’s nothing. To paraphrase someone who paraphrased Patrick McKenzie:
Patio11's Law: The internet is bigger than you think, even if you take this law into account.
13k? That’s minuscule!
Even considering that you, reading this, are probably an outlier, a weirdo, an obsessive learner, a compulsively curious spiky personality — there’s still gotta be at least 1 million more people like you and me who understand English out there.
That’s 1.3% market penetration. 🚢
But realistically, there’s quite a bit more than 1 million people like us roaming this rock.
I just gotta keep plugging at it and find our tribe.
The internet allows you to be an emitting antenna, but also a lighthouse to try to attract the right people who dig your stuff and like similar things. Be the magnet that attracts the needles from the haystack. Get the band back together, as Jake and Elwood would say.
That journey sounds like a good way to have fun, learn, explore and create.
🛀 Investing is largely psychology, not math.
I don’t mean “you have to predict what others will do based on sentiment”, I mean “you have to know and control yourself” because even if you have the math right, if you do the wrong things, or can’t bring yourself to do the right things, you’ll still lose.
Sometimes I think about how many people I've talked to because of the internet, and feel like I may have to hand back my introvert card...
🤔 I haven’t heard much about universal basic income (UBI) in a while.
💚 🥃 💃 If you want to become a paid supporter but every time you click on the subscribe link you don’t see options to do so, it’s because you’re not logged in your Substack account on that browser/device.
Liberty’s Highlights is reader-supported. To support my work, consider becoming a paid supporter. 💃
🧩 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 💳 💴
🗜💸 Amazon’s Belt-Tightening 🚚 📦📦📦📦
Andy Jassy is leading a cost-cutting review of the tech giant and paring back on businesses at the company that haven’t been profitable, according to people familiar with the matter. [...]
As part of the monthslong cost-cutting review, Amazon has told employees in certain unprofitable divisions to look for jobs elsewhere in the company, because the teams they were working on were being suspended or closed [...]
The same story that the Stripe and Shopify founders wrote about in their letters applies to Amazon:
Between the end of 2019 and end of 2021, Amazon hired more than 800,000 employees, mostly at its hundreds of warehouses, as it sought to keep up with a surge in online orders.
We have to remember that investments have long lead time. Building infrastructure and hiring people isn’t like an investor hitting the “buy” or “sell” button, there’s a delay of many months — a lot of decision should be kept in context of when they were made.
What’s important is constantly updating and fixing mistakes when they become apparent.
Amazon’s leadership is closely evaluating its Alexa business, according to some of the people. The business has more than 10,000 employees and is a major recipient of investment capital, some of the people said. Internal documents viewed by The Wall Street Journal show that in some recent years Amazon’s devices unit, which includes Alexa, had an operating loss of more than $5 billion a year.
Not a bad idea to review investment in Alexa, because while it’s a good space to be in, it turned out to be a *smaller* part of people’s lives than was believed a few years ago when the narrative was that voice would be one of the big “pillars” of people’s digital lives, like smartphones, TVs, and laptops…
It just didn’t work out that way.
Echos are cool to check the weather, play music, and set timers, but people don’t do tons of elaborate searches for information on them, and they certainly don’t do much shopping on them.
Now there are no doubt indirect benefits to Amazon’s investments in the space, like gaining a lot of machine-learning expertise that can be useful in other areas of the company (especially within AWS).
But there may be diminishing returns to having that much headcount working on Alexa...
Amazon last underwent an extensive profitability push in 2017 under Mr. Bezos. Senior Amazon executives say Mr. Jassy’s review is much more extensive. Mr. Jassy has been laser-focused on profits since taking over last year, and has presided over tough decisions, the people said.
What was the result of the 2017 push?
While you can’t attribute results to *one* thing, it certainly seems like FCF/share inflected after that. It’ll be interesting to see what happens as the effects of the current belt-tightening start to be felt in the next couple of years.
📉💣 Another victim of FTX’s demise: Philanthropy 💸
I didn’t know that Sam Bankman-Fried is (was) one of the biggest donors to ‘effective altruism' causes.
Mr. MacAskill argues that people living today have a responsibility not just to people halfway around the world but also those in future generations.
The rise of this kind of thinking, known as longtermism, has meant the Effective Altruists are increasingly associated with causes that have the ring of science fiction to them — like preventing artificial intelligence from running amok or sending people to distant planets to increase our chances of survival as a species.
Mr. Bankman-Fried makes his donations through the FTX Foundation, which has given away $140 million, of which $90 million has gone through the group’s Future Fund toward long-term causes. [...]
Mr. Bankman-Fried said he expected to give away the bulk of his fortune in the next 10 to 20 years. (Source)
Well, I’m guessing those cheques will be a lot smaller now 😬 which is too bad for those good long-term, high-risk/reward projects that may not get much attention from traditional philanthropy.
Update: As I’m about to post this, this headline came out:
Sam Bankman-Fried steps down as FTX CEO as his crypto exchange files for bankruptcy
TSMC looking at additional gigantic 3nm fab in Arizona..?
The scale of the investment is expected to be roughly similar to the $12 billion it committed two years ago, the people said. [...]
TSMC’s new facility would manufacture so-called 3-nanometer transistors
The journal wrote a couple pages on this, but those 2 lines are basically the meat of it.
This would be in addition to the fab they’re already building in AZ.
Interesting how in the US, so many extremely water-hungry fabs end up 🌵 in the desert 🌵 ¯\_(ツ)_/¯
🇪🇺 EU wants to build its own satellite constellation to beam down internet access 📡 🛰🛰🛰🛰🛰🛰🛰🛰🛰
EU countries and European Parliament lawmakers are likely to reach a deal on a 6-billion-euro ($6 billion) satellite internet system [...]
First, how convenient is it that Euros and GBPs are now basically 1:1 with the USD?
A space-based network could back up terrestrial networks in the event of major outages or disasters, and offer connections in places not covered by traditional service providers. [...]
The proposed satellite internet system could lead to the construction and launch of up to 170 low orbit satellites between 2025 and 2027.
The companies active in this area include Elon Musk's SpaceX, Amazon' Kuiper Systems and British satellite company OneWeb. China also has its own constellation project.
European players include France's Eutelsat Communications, the world's third-biggest satellite operator by revenue which is merging with Britain's OneWeb, Arianespace and Thales Alenia Space. (Source)
170 sats means this isn’t going to be anything like Starlink, with its 3000+ satellites (with more on the way).
Sounds like it would truly be just a backup plan and one more example of each country and region trying to become less dependent on others (which has pros and cons, to be sure).
VCs are sitting on half a trillion dollars? 🤨
🧪🔬 Liberty Labs 🧬 🔭
Google 💔 JPEG XL 😢
I’m a little mad at Google. The Chrome team in particular.
As an efficiency-geek, I’ve been excited to see progress on the adoption of the JPEG XL format. It’s a huge improvement over the ubiquitous JPEG that was created in 1992 (we’re overdue for an upgrade).
I wrote about why this tech is so cool back in edition #154 (direct link to the relevant section). It truly is the Swiss Army Knife of image compression formats.
But the Chrome team recently decided to do this:
Thank you everyone for your comments and feedback regarding JPEG XL. We will be removing the JPEG XL code and flag from Chromium for the following reasons:
- Experimental flags and code should not remain indefinitely
- There is not enough interest from the entire ecosystem to continue experimenting with JPEG XL
- The new image format does not bring sufficient incremental benefits over existing formats to warrant enabling it by default
- By removing the flag and the code in M110, it reduces the maintenance burden and allows us to focus on improving existing formats in Chrome
Since Chrome has the biggest browser market share (and others like MS Edge are based on the Chromium engine), this is a serious hindrance to the adoption of the new format.
This piece explains why Google should rethink things and support JPEG XL. Here’s a highlight:
In the past, new image formats have been introduced that brought improvements in some areas while also introducing regressions in others.
For example, PNG was a great improvement over GIF, except that it did not support animation. WebP brought compression improvements over JPEG in the low to medium fidelity range but at the cost of losing progressive decoding and high-fidelity 4:4:4 encoding. AVIF improved compression further, but at the cost of both progressive decoding and deployable encoders.
We looked at six aspects of JPEG XL where it brings significant benefits over existing image formats:
Lossless JPEG recompression
Lossless compression performance
Lossy compression performance
Works across the workflow
The reader can judge for themselves if they consider these benefits sufficient. In my opinion, every one of these benefits is sufficient. Most importantly, JPEG XL can bring these benefits without introducing a regression in other areas, at least in terms of technical strengths.
Come on Google, do the right thing!
☀️🧠 Vitamin D and Brain Aging
From the abstract of a new study:
Structural MRI data and vitamin D levels were obtained in 1,865 subjects from the general population. Linear regressions were applied to investigate the association of vitamin D levels and vitamin D deficiency with imaging derived brain age, total brain, gray matter and hippocampal volumes. Different sets of covariates were included.
Vitamin D deficiency was significantly associated with increased brain age. Also, linear vitamin D levels were significantly associated with total brain and gray matter volumes, while no significant association with hippocampal volume was found. Further interaction analyses showed that this association was only significant for male subjects.
Our results support previous findings suggesting that vitamin D-deficient individuals have an accelerated brain aging. In addition, associations between vitamin D levels and total brain/ gray matter volumes suggest neuroprotective effects of vitamin D on the brain.
While vit D is not magical, it seems to be pretty asymmetric in that there’s little toxicity to it even in high doses (our bodies have evolved to deal with massive doses back when we spent every day outside, which few of us do now), and it seems to have all kinds of health benefits, closer to a hormone than a vitamin.
I’ll keep taking my 5,000 to 10,000 ui/day depending on the season (that’s for a 6’1” male living in Canada). Note that it’s better absorbed with large meals and taken as gelcaps (because it’s fat-soluble).
🎨 🎭 Liberty Studio 👩🎨 🎥
🗣🎙 Interview with the founder of Midjourney 🎨🤖
Ben Thompson (💚 🥃 🎩) has a really good interview with David Holz, the founder of Midjourney (and previously Leap Motion). I recommend the whole thing:
An Interview with Midjourney Founder David Holz about Generative AI, VR, and Silicon Valley (there’s also an audio version 🎧)
Here’s an interesting highlight about the importance of creating a community around generative AI art:
Q: you don’t have to ask to feel stupid and say, “How do I do this?” You literally just sit and observe, which is how humans learn, generally.
A: When I did user testing, it was kind of unbelievable. It’s like, “Don’t you want a person to discover the product by themselves?” We would do this and we’d be like, “Okay, here’s a machine. It’ll let you do a picture of anything you want, anything you can imagine, what do you want?” And they just go, “Dog.”
And it’ll show them a photo of a dog and they go, “Okay.” And it’s like, “Well no, come on..” Because you’re there at first. “What do you want? Come on a little bit more than that.” And they go, “Big dog.” And then I keep questioning and they go, “Big fluffy dog.” And at the end of it they’re so uninterested, it’s like, “This isn’t interesting, why would I care about this?”
But then you throw these people into the same environment all of a sudden, with complete strangers, they go, “Dog.” And someone else goes, “Space dog”, “Space dog with lasers”, “Space dog with lasers and angel wings”. And all of a sudden this person’s like, “Oh my god.”
This has been my experience too — I find these tools so interesting *because* I can see what others are creating and how they are creating it.
Midjourney has prompts shared by default in the public Discord channels, and there are communities elsewhere where people will often share their prompts — ArtHub.ai is neat, and Reddit has a bunch of communities.
Here’s my Reddit account where I’ve been posting some of my stuff.
The ability to iterate on what others are doing, for the cream to rise to the top, and for the really effective ways-to-poke-the-machine-to-get-it-to-do-something-cool to spread and be further refined is creating a tight feedback loop that generates much faster progress and nicer art than if everyone was doing it separately.
That’s on the human creative process.
In the technical, compute infrastructure side, this part about the scale and growth of MJ is fascinating:
Q: You’re using thousands and thousands and thousands of GPUs.
DH: More than thousands. Or usually more than 10,000 GPUs.
Q: Are you the largest GPU user in the world?
DH: No, but we are one of them. [...]
So right now, if you make an image, there are eight different regions of the world that the image might get made in, and you have no idea. It might get made in Korea or Japan or the Netherlands or something. It’s going to eight different regions that the GPUs are balancing between. Where it’s really cool is because we’ll use a lot of the GPUs in Korea while it’s nighttime and everyone’s sleeping there, no one’s using them. We can kind of load balance. You can basically race the darkness of the night across the earth. [...]
almost nobody realizes that we’re effectively already within a factor of 10 of the world running out of GPUs for this kind of market. So hopefully in the next year we go up a factor of 10 otherwise the cloud just runs out of machines, which will be a really interesting thing to happen. [...]
what’s happened is there wasn’t really that much consumer demand for compute, and now we’re showing that there is, and the clouds are basically not built out under the assumption that there’s a large consumer demand for compute.
Basically the cloud will have to readjust to this new reality. And that is a pretty big readjustment. To me, if I had to guess, if we want a billion people in the world using large AI models, whether for AI image generation or for text or for anything, these are just large models, it doesn’t really matter what they’re being used for. If we want a billion people to have access to it, the cloud probably needs to be a thousand times larger computationally speaking than it is today. And so a thousand times difference, it’s like it’s not physically possible in the short term.