185: Satya Nadella's Strangest Deal, Daniel Ek, Cloudflare in the Basement, Apple A15, OSAM, Ho Nam, AI Model Size, ASML Investor Day, and Chronobiology

"in August 2020 there was a deeply surreal episode"

The greatest remedy for anger is delay.


😴 🛏 Is the position in which you sleep just habit, or hereditary?

People have so many ways and combos that they seem to be pretty consistent about… Side, back, belly, hugging a pillow, etc.

I’m curious if this has ever been studied… Do kids sleep in postures similar to their parents? What if each parent sleeps differently, are some sleep postures recessive and others dominant (to use genetic terminology)?

Can you consciously learn a new sleep posture habit and keep it, or is your body unconsciously always going to try to go back to whatever you first came to naturally?

I tend to sleep about 75% on my back and 25% on the sides, which is why I got one of these pillows:

If you’re a back/sides combo sleeper like I am, I really recommend trying, it was a game-changer for me. I used to constantly have a crick in the neck, but not anymore!

🔊 I’ve had my new AirdPods Pro for a few days now.

One thing that is slightly worse than with regular AirPods is eating. Because the Pros seal your ears more fully, when you chew your food, you can kind of hear that sound resonating inside your ears. It’s hard to describe, but it’s louder than with the regular AirPods. It’s a bit like if you plug your ears with your fingers and then talk or chew, it’s going to sound different.

On noise-cancelling: Maybe I’ve primed myself lately to look for great technologies that we take for granted and don’t notice anymore, but this stuff could basically pass off as magic if you could travel back in time and demonstrate it to someone. It’s so effective that when I put them on, I sometimes feel like there’s just been a power-failure in my house because the noise from the fridge and furnace fan stops cold.

I’ve never had noise-cancelling headphones before, because I use speakers for music (I’ve always worked from home) and small portable earbuds for podcasts/audiobooks/music on the go.

I knew the principle of the inverted soundwaves, and had heard descriptions of the effect, so I knew what to expect, but it’s still very cool to experience it. Our non-scientific, intuitive brain feels like noise isn’t something you can make disappear.

I think it’s probably even more impressive with tiny buds like the AirPods Pro than with big, over-the-ear headphones, because with those your brain can kind of assume that a lot of the noise-blocking comes from the physical barrier, while with these tiny things, it feels more like pure magic when the background fades away.

Transparency mode, which brings a lot of the sound back, also feels pretty magical, and is very practical when you want more situational awareness.

💚 🥃 How much quality content does a guy have to give out before he gets a few bucks in the tip jar?

💙 Subscribe now 💙

Investing & Business

‘Strangest thing I’ve ever worked on,’ says Satya Nadella

Just by seeing that headline, I’m sure you can guess what it refers to…

It’s difficult to remember now, but in August 2020 there was a deeply surreal episode wherein Microsoft proposed acquiring some portion of TikTok in a complicated deal as the Trump administration was threatening to come down hard on the platform. Ultimately, TikTok turned Microsoft down just a day before Oracle came in with an entirely different kind of deal and the whole thing dissipated.

Speaking today at the Code Conference, Microsoft CEO Satya Nadella said the whole ordeal was the “strangest thing I’ve ever sort of worked on.”

In terms of the value of such a deal for Microsoft, Nadella said he “was kind of intrigued by it,” before trying to move right along with a laugh, deadpanning after a pause “and then I guess the rest is history.” [...]

“TikTok came to us, we didn’t come to TikTok,” he says. [...]

“There was a period of time when I though the the [US government] had a particular set of requirements, [but] it just disappeared,” Nadella says. “President Trump I think had a particular point of view of what he was trying to get done ...and then just dropped off.” (Source)

Franklin Templeton to Acquire O’Shaughnessy Asset Management & Canvas Software Platform

This def qualifies as big news in our circles! congrats to Jim (💚 🥃) & Patrick, I’m sure building OSAM has been quite a journey (not that it’s over, but this is a big milestone).

OSAM’s capabilities, both as a factor-based investment manager and as a Custom Indexing solution via OSAM’s flagship Canvas platform, will serve as an important expansion and enhancement of Franklin Templeton’s existing strengths in SMA and custom solutions capabilities. The Canvas platform launched in late 2019 and has seen strong growth since its inception, now representing $1.8 billion of OSAM’s total $6.4 billion in assets under management as of August 31, 2021. [...]

“Custom Indexing represents the next progression of investing through Indexing, ETFs, and Direct Indexing,” said Patrick O’Shaughnessy, CFA, Chief Executive Officer of OSAM.

“The financial terms of the deal weren’t disclosed.”

Tip of the hat to all involved, including all the 40+ employees who may not be as publicly visible, but have helped build the firm 🍻

Interview: Daniel Ek, Spotify co-Founder & CEO

I really enjoyed this one, especially the part about the challenge of fixing problems int he world of “atoms” (as opposed to the digital world of “bits”).

Sometimes, I almost think that the main way this can work, at least in the private sector, is for those who made fortunes in the digital world to leverage that access to capital and talent and that credibility to start doing “atoms” companies or philanthropy.

Once again, big congrats to Patrick O’Shaughnessy on five years of making people and ideas more accessible to a whole ecosystem, and raising the level of the discussion in the community.

‘by my definition, a great company doesn’t need your $’

Ho Nam dropping some wisdom:

In VC, extreme concentration of $ in small # of assets happen as a consequence of the power law - due to value appreciation rather than invested $. Extra alpha may be found by applying more $ in rare situations where value >> price. Or prior gains will be frittered away if wrong.

The most important thing is to have a quality asset (and before that, to build a quality company). Investing more dollars is secondary to continuing to build the company and compounding $ already deployed. No need to apply more $ (should be the exception rather than the rule).

There is plenty of $ in this world. And plenty of “helpers” to advise, manage, etc. But not enough good places to deploy $ (by my definition, a great company doesn’t need your $). And even harder to find a seller of an amazing asset at a good enough price.

For more, check out the interview I did with Ho Nam.

“Long-Distance” Calls — Where are Regulators?

It’s not really something I want to dig into because it’s boring, but a quick search revealed to me that telecos still appear to have the concept of “long-distance calls”, at least here in Canada and the US.

I know a lot of packages have “unlimited XYZ” or whatever, but you shouldn’t have to pay more or “upgrade” or have market segmentation based on this…

It’s 2021, I can watch 4k HDR video streaming from a server on the other side of the world, transferring multiple gigabytes of data, for the flat-rate I pay my ISP. Most phone calls are probably under 1 megabyte (modern codecs are really good at compressing a mono stream of voice — it’s not like you need great dynamic range or fidelity to keep voice clear enough).

The concept of a “long-distance call” should long ago have been fully abolished.

If regulators want to do something that is pro-consumer, this is the thing to do. A neat little legislation, it can be 1-page long, that says “all phone calls should be charged the same rate regardless of distance”.

It’s all bits going through the same pipes at this point…

Cloudflare Rolling Out in… Office Buildings Basements

Ambition x innovation:

Introducing Cloudflare for Offices. We are creating strategic partnerships that will enable us to extend Cloudflare's network into over 1,000 of the world's busiest office buildings and multi-dwelling units. These buildings span the globe, and are where millions of people work every day; now, they’re going to be microseconds away from our global network. Our first deployments will include 30 Hudson Yards, 4 Times Square, and 520 Madison in New York; Willis Tower in Chicago; John Hancock Tower in Boston; and the Embarcadero Center and Salesforce Tower in San Francisco. [...]

Cloudflare has built a mutually beneficial relationship with the world's ISPs by reducing their operational costs and improving customer performance. Similarly, we expect a mutually beneficial relationship as we roll out Cloudflare for Offices. Real estate operators & service offices upgraded with this amenity increase the value and occupancy of their portfolio.

This is like the Cloudflare network going full fractal.

Hyperscalers have these massive datacenters that cover whole regions of the world.

Cloudflare has a more distributed footprint, now with around 250 points of presence in cities around the world, so they are close to most of the world’s populations (often inside the same buildings as the local ISPs).

But now, they’re taking this one step further and atomizing the location of their equipment down to the office building level… Reminds me a little of the Google search appliance.

Cloudflare for Offices simplifies this: a direct connection to Cloudflare’s network puts all office traffic behind Cloudflare’s services. Now, creating an office is as simple as plugging a cable into our box, and all the security and performance features that an office typically needs are microseconds away. [...]

Cloudflare for Offices is the onramp for offices onto Cloudflare One. It’s a fast, private onramp for your office network traffic straight onto the Cloudflare network [...]

Having Cloudflare equipment close means that the performance of the user device will be vastly increased as opposed to having to connect to a far off data center. [...] imagine what happens when you’re connecting to Cloudflare in your office basement.

This is hard work because it doesn’t scale as easily as a more centralized structure, but because it is hard work, it probably means that there won’t be a ton of competition trying to get deals in every city and every big office complex, and once Cloudflare is in, it probably reduces the appetite of these property managers from doing other deals, at least a little.

Speaking of Cloudflare, here’s a bonus joke speculating on why they named their recent storage service R2:

And if you want even more, Ben Thompson (💚 🥃 👨‍🎓) has a great free piece about R2 and how the company is disruptive and trying to become the 4th big cloud.

💙 Subscribe now 💙

Science & Technology

Apple’s New A15 Chip & 🍎’s Custom Silicon History

René Ritchie, one of the very best Apple-watchers, with the technical chops to understand the engineering and the communication skills to explain it simply, has a good video giving an overview of Apple’s latest slice of silicon (found in the iPhone 13).

The video includes a brief history of all of Apple’s mobile chips, from the very first iPhone to today, and even if you’re not a transistorhead, I think it provides a great overview of the trend and how much has been accomplished in a relatively short amount of time.

Interview: Dr. Satchin Panda, Chronobiology Edition

Good interview with Dr. Panda. I’ve heard a few of his interviews over the years, and they’ve generally been good and worth the listen.

In this one, he spends the first part on various aspects of sleep and our circadian rhythm, and in the second part, goes more into diet and time-restricted feeding/intermittent fasting:

You can get the audio-only podcast version at the link above, and there’s also a transcript.

🧙‍♂️ ASML Investor Day 🔮

litho etch litho etch’ on Twitter has a nice thread recapping some of the highlights from ASML’s recent investor day.

Here’s a couple that stood out to me, but check the thread, or the whole investor day for more:

ASML sees litho-density scaling thru at least 2030 and TSMC projects a roadmap for semiconductor scaling out to 2040. Continued semiconductor scaling is great for society!

Another amazing slide showing how ASML's installed base pumps out more wafers every year. An underappreciated aspect of how semi fabs are able to drive costs down year after year and why we can expect the EUV per wafer cost to keep coming down as the tech matures.

h/t friend-of-the-show and supporter (💎🐕) Rishi Gosalia

A.I. Model Inflation

From TheSequence:

It seems that every other month we have a new milestone in the race of building massively large transformer models. The trajectory is astonishing. GPT-2 set up new records by building a 1.5 billion parameters model just to be surpassed by Microsoft’s Turing NLG with 17 billion parameters. GPT-3 set up the mark at 175 billion parameters, and Google’s Switch Transformer took it to 1.6 trillion parameters. Recently, the Beijing Academy of Artificial Intelligence (BAAI) announced the release of Wu Dao 2.0, a transformer model that contains a mind-blowing 1.75 trillion parameters. Those numbers are just hard to imagine. 

We’re now in the era of trillion-parameter models.

Reminds me of how growing up, CPU clock speeds were all megahertz, and then suddenly, we switched over to gigahertz. What used to be a big deal (an extra 100mhz) over time becomes a rounding error in the new scale…

Clock speeds have since stagnated because the thermals just don’t make sense passed a certain point, and it’s easier to get more performance by widening pipelines and adding more cores and cache and specialized chip areas (ie. to decode h.265 video or whatever).

I wonder if/when we’ll get to diminishing returns with these AI models, or if they’ll just keep scaling as long as we can throw the compute and memory at them (100 trillion parameters? 500 trillion?).

At some point do you just run out of data to feed the model, or will this problem solve itself since by the time we get there, we’ll be producing so much more useable data — including all the data produced by previous AI models — that the machine will stay well-fed for the foreseeable future..?

🥶 Cold Data Storage in DNA, Follow-Up Edition 🧬

I wrote a little about this in edition #184. I also liked this quote about it:

you get DNA storage with a very light footprint.

I mean I'm sure you can store all the data of the year [on the] internet in that mug.

So you just think of the density and then longevity so you can replicate the data in terms of longevity is tens of thousands of years.

Seems like Hari Seldon should create his Foundation knowledge repository with DNA…

The Arts & History

Everything is a Remix, Army Drill Sergeant Edition

I’ve posted a few times before about how the internet, and especially platforms like Tiktok, allow these permissionless collaborations between people who would never ever meet or get together otherwise, often without one party even knowing that the other party is creating something using their stuff. (see the bottom of editions #163 and #177)

Well, I’m pretty sure that Drill Sergeant Depalo didn’t see this one coming, but I think the result of this duet with The Kiffness is pretty cool.

It’s giving me some Full Metal Jacket flashbacks.

This is the original song by Depalo. He’s based in Fort Jackson, South-Carolina.