96: Steve Jobs AI-in-a-box, Nvidia Q4, Twitter Super-Followers, John Malone, Software's Teeth, TSS M&A at Constellation, and Mad Men with Smartphones

"Are AWS and Azure the ventricles of the heart?"

Let no one be deluded that a knowledge of the path can substitute for putting one foot in front of the other.

—Mary Caroline Richards, poet, potter, and writer best known for her book Centering: in Pottery, Poetry and the Person.

So a tweet I wrote about both Bill Gates and Bitcoin went kind of viral (a few thousands of likes and hundreds of RTs). Wow, what was I thinking?

I almost deleted it just so the flood of stupidity in my mentions would stop (I ended up muting the thread)…

It’s a bit demoralizing to see that nobody actually reads or tries to understand what is said, they just use the opportunity to reiterate whatever they believed going in: ie. The bitcoin bulls all say Gates is dumb or evil or just saying that because he’s rich, and the bitcoin bears all see this as Gates agreeing with them; meanwhile, Gates actually said that his focus is trying to solve big problems in the world and he’s investing in ways that help accomplish that, so he’s not interested in speculating on the price movement of something — would the world be a better place if Gates and Buffett were putting their billions in crypto coins, or as they do, in labs working on malaria and HIV drugs and K12 education and clean water for the poorest people on Earth and clean energy R&D? — seems pretty clear to me, but hey ¯\_(ツ)_/¯.

I guess it’s the weird dichotomy of Twitter. Building an audience can be fun, as you get to talk with more interesting people over time. Going viral is terrible (I can’t imagine how it is for people who go truly viral and have thousands of RTs and hundreds of thousands of likes on a Tweet).

It was a good reminder of how nice the little twitter bubble that I usually spend my days in actually is. We can complain about it all we want, but compared to the mouth-breathers that populate large parts of the net, it’s really cozy.

🛀 In edition #95 I featured a visual search engine that finds images that have the “same energy” to either an image that you select or uploaded, or based on keywords.

It was created by Jacob Jackson (formerly of OpenAI, which is always nice to have on a resumé — do people even have resumés now? “On your LinkedIn profile” doesn’t roll off the tongue as smoothly..).

Anyway, I had a nice chat with Jacob, and we got on the tangent of AIs exhibiting what would be considered “taste” by humans.

I had a riff on this in edition #3 (one that you’ve only read if you’re an OG of this show — I really recommend you click that link and look at that great portrait of a lady with an axe, I love it):

I’ve always wondered how much an aesthetic sense can be trained, and how much is innate. If machines can eventually become good at what humans would describe as “having taste” without relying on meta-data about “quality” that comes from humans (as Archillect does, directly and indirectly), that may teach us something about ourselves.

Jacob and I had some back-and-forth about this… Basically, the idea of “taste” in itself seems pretty intractable, kind of like consciousness, partly because it’s so hard to define; but I’m thinking that as long as you have an output that is something that most people would agree is “good taste”, you could still get somewhere even if you don’t understand how it all works mechanistically (which is the case of a lot of AI/ML black boxes).

I wrote: “if some AI becomes an editor, a bit like Steve Jobs, and everybody agrees that this AI has great taste, what a tool that would be”

And Jacob suggested that GPT-3 may be an example of an AI having taste in writing. It can't tell you whether something is good or bad, but it knows how to continue bad writing with more bad writing, and good writing with more good writing — so the knowledge exists, even if it can't name it.

So I’m thinking, what if over time, as these models iterate and the hit/miss ratios become much better on the creation side (let’s imagine what GPT-7 maybe be able to do, for example), what if they could be trained to be the judge of something, rather than a creator.

What if you could have an AI that is generally regarded as having great taste, being a great editor, a kind of Steve-Jobs-in-a-box..?

What kind of interesting things could we do with AIs like that, and how would that change the world (since they can scale to almost any amount of tasks, be copied and modified at will, work 24/7 at inhuman speed, etc).

Taste is such a bottleneck to making a better world. We don’t often talk about it, but we’re seeing how scarce things like taste and ambition are when people who have super-human amount of these things (combined with other qualities, of course) reach escale velocity and change the world (there’s survivorship bias, but this is a case of necessary-but-not-sufficient, so it still matters a lot). Anyway…

This wouldn’t be the show without a dad joke, so here’s me to Jacob after he replied something good:


Investing & Business

Nvidia Q4

-Overall revenue: +61%
-Data center revenue: +97% (wow)
-Gaming rev: +67%
Gross margin FY: 65.6% (+310 bps)

And that's while being supply-constrained on a lot of stuff... imagine if they could sell as many things as there was demand for...

From the call, here’s Jensen:

at the company level, we're supply-constrained, demand is greater than supply. And we have enough supply. We usually have enough supply to achieve better than the outlook. And we had that situation in Q4. We expect that situation in Q1. And we have enough supply to grow through the year.

If that’s not telegraphing that they’ll beat guidance, I’m not sure what is…

Overall, pretty bonker results at that scale. They added almost 6 billion of revenue for their fiscal '21.

Full-year revenue for the data-center segment was $6.70 billion, up 124 percent.

Unless trajectories change pretty drastically, they will soon make more from data-centers than from gaming, the segment on which the company was built (TTM: Gaming full-year revenue was $7.76 billion, up 41 percent).

And all this is with the automotive segment down 23% for the full-year, because of the pandemic. When that starts growing again, it’ll kick things up another notch (especially as more and more car models start shipping with autonomy and AR features — or at least, advanced safety features that use some of the same sensors and software) and could, over time, grow into another important pillar.

Some interesting stuff on Mellanox/BlueField DPUs:

I believe that every single data center node will be outfitted with a DPU someday. And that someday is probably, call it, 5 years from now. And the fundamental driver of it is going to be security. Every single application in the data center and every single node in the data center has to be individually secured. Zero-trust Computing… these initiatives are going to cause every data center to have every single application and every single node be secured. [...]

So every node will be a DPU process for the software, for the infrastructure. You're essentially going to see the data center infrastructure be offloaded from the application plan, and it will be something like a BlueField. So I think this is our next multibillion-dollar opportunity

Ok, that's a good line by Ben Thompson

In the context of a post about the automotive chip shortages ($ sub. required):

if software is eating the world, then [semiconductor] chips are its teeth

Now there’s an image. 🦷🦷🦷🦷🦷🦷

I kind of wish I had come up with it… (Are subsea fiber-optics cables arteries? Are AWS and Azure the ventricles of the heart? Charter and Comcast are the capillaries? GPS is the inner ear? Spam is fingernails..?)

Scuttleblurb Writeup on Twitter

Twitter is having kind of a moment lately, after years of stagnation on the product and financial side. Friend of the show David Kim of Scuttleblurb (𝕊𝕡𝕖𝕔𝕚𝕒𝕝 𝔼𝕕𝕚𝕥𝕚𝕠𝕟 #𝟙 interview) did a post on Twitter ($ sub. required), check it out:

Speaking of that interview with David, it was nice to see this.

Twitter’s Super-Followers & Other Experimentations

So there’s going to be “super-followers” on Twitter soon. Did the Twitter product people spend too much time on Tinder during lockdown?

I just hope it's mostly like tipping to support people who you like/are getting value from, and doesn't massively fragment the timeline with everybody having their own paywall, and sub-communities becoming balkanized between the rich heavy-spenders who can see everything and the poor people stuck as second-class citizens on the free feed that is full of “teasers” for the pay sections.

And my worry isn’t that everybody will suddenly start putting stuff on their premium feed, but that because of the power-law nature of things in social networks, even if just 1% of users go premium and paywalled, it'll probably be like 90% of the best accounts that contribute the most, and that would make a difference.

So that’s a bit worrying, but I’ll be waiting to see more details before judging it too much… 👩🏻‍⚖️

I’m not against the idea, and I’m not above sending or accepting money on Twitter. But again, things aren’t binary, and there can be things I worry about even when something seems like a good idea.

TSS M&A Pre-and-Post Acquisition by Constellation Software

Interview: Tren Griffin

I enjoyed this interview of Tren Griffin by Howard Lindzon. It’s not about anything super-specific, more about Tren’s life and general philosophy. But I'll take it, because Tren recharges my joy-of-living and curiosity batteries every time I hear him speak.

He's got that smile on his voice and excitement in his speaking cadence… We all need and injection of that straight in our veins once in a while, so here it is:

Interview: John Malone (60-min podcast)

Good hour-long conversation with the doc, covering a bunch of stuff:

Bill Gates came to see me with a six pack of beer and a pizza when negotiating a contract. He told me one time he said, John, forget about hardware. There's no piece of hardware that can't be emulated in software. And damn, I failed to listen to it. [...]

My dad always said: First, whatever the problem is, try and guess at the answer.

Try and develop your instinct.

Then use the techniques they teach you in school.

Over time you will train your intuition and probably guess the answer better than the technical method."

h/t Frederik ‘Neckar’


Science & Technology

‘What are the drivers of deforestation?’

Good new post on Our World In Data by Hannah Ritchie:

Since the turn of the millennium, the world has been losing around 5 million hectares of forest every year. Nearly all of this occurs in the tropics; almost half of all deforestation takes place in Brazil and Indonesia.

Three-quarters is driven by agriculture. Beef production is responsible for 41% of deforestation; palm oil and soybeans account for another 18%; and logging for paper and wood across the tropics, another 13%. These industries are also dominant in a few key countries.

Effective solutions will be focused on these agricultural activities and those countries where most deforestation occurs.

One more way in which synthetic meat will be great when the tech matures.

(We still need a good name for it.. nothing with “lab” or “grown” in it. I floated “clean meat” a while ago, but I’m not sure how much I like it anymore. “Green meat” certainly doesn’t sound appetizing, even if you don’t take into account the Soylent Green undertones…).

Interview: Josh Waitzkin

Not sure in what category to put it, but I enjoyed this conversation between two friends who happen to by Josh Waitzkin and Tim Ferriss.

The ‘what advice would your future self give you?’ mental pump-primer is very interesting. What would he tell me? 🤔

I read Josh’s book, ‘The Art of Learning’, many years ago and really enjoyed it. I went in only knowing that he was the chess prodigy that the film ‘Searching for Bobby Fischer’ is based on, but it turned out to be about so much more than that (I learned a lot about push hands martial arts…), but mostly, just about an interesting mind who has developed some very good techniques to be an effective learning machine.

Clearly the ability to learn is the most important meta-skill, because all the others are built on it. So learning from people like Josh is probably a good idea.

Not that the interview above is so focused on that, but you can still get an idea for how he thinks. If the interview above (or the first one he did with Ferriss) intrigues you, I recommend the book.


The Arts & History

This One Just Tickled Me ¯\_(ツ)_/¯

Dear Diary: today i got [paid] to draw Doomguy teaching Rapunzel, from Disney's "Tangled", how to shoot the Heavy Assault Rifle

today was a good day

Source.

Mad Men with Smartphones

So I've been rewatching Mad Men lately, and the way they constantly reach for cigarettes and hard liquor reminds me of how we reach for our smartphones today.

Amadeus GIFs

So I decided there weren't enough high-quality GIFs of Amadeus out there... Made a few to add to my GIF conglomerate.

Previous ventures include:

The Royal Tenenbaums GIFs,  Deadwood GIFs, Mad Men GIFs