SITALWeek #449
Welcome to Stuff I Thought About Last Week, a personal collection of topics on tech, innovation, science, the digital economic transition, the finance industry, and whatever else made me think last week.
Click HERE to SIGN UP for SITALWeek’s Sunday Email.
In today’s post: entrepreneurs are boomeranging back to big tech, leaving open the question of who will control the application layer for AI; AI learns to reason; Waymo is a very good driver; Microsoft achieves in error-corrected qubits; wireless data growth surprises in the US; Minecraft LLM players; AI gets a wallet; interesting Alzheimer's data; Nick Cave updates his views on AI; Stephen Wolfram explains the computational irreducibility of LLMs; CEOs have become boring; and, much more below.
Stuff about Innovation and Technology
Wireless Is More
Wireless data usage grew 36% in the US in 2023 and is poised to expand further with growing AI use cases. The number of connected devices also grew by 6% over the prior year. All in, Americans used a record 100 trillion megabytes of mobile data.
Error-Corrected Qubits
Microsoft researchers have been able to use 56 qubits to achieve 16 logical qubits. Logical qubits are achieved with error correction that spreads out the impact of mistakes inherent to the nature of quantum mechanics. Microsoft is striving for 100 logical qubits. Although a timeline is uncertain, they have tripled the logical qubits in one system since April.
Way Mo’ Safe
Waymo released a new safety dashboard on its autonomous vehicles. With over twenty million ride miles (through June 2024) of data, Waymo had 84% fewer airbag deployments, 73% fewer injury-causing accidents, and 48% fewer police-reported crashes compared to human drivers. Further, of the 23 most severe incidents, 16 were caused by a human driver rear-ending a Waymo and 3 were caused by human drivers running red lights.
Digital Agent Microcosm
Startup Altera created a Minecraft world with over 1,000 interacting agents all powered by LLMs. The agents formed their own democracies and followed their dreams. The priest ended up trading the most because he was bribing townspeople to convert. As I’ve noted many times, I expect the inevitable agent-based economy will dwarf our analog economy in size and scope, and it’s likely to produce the majority of innovation and scientific breakthroughs in the future. In related news, Coinbase has had its first AI-to-AI transaction using crypto, the likely currency of the agent economy.
Sugar-Starved Neurons
IDO1 inhibitors appear to improve brain metabolism with the potential to reverse Alzheimer’s cognitive decline in mice. Drugs that target IDO1 are already in clinical trials for cancer treatment, so they could be used to measure the impact on Alzheimer’s in humans as well.
“The kynurenine pathway is over activated in astrocytes, a critical cell type that metabolically supports neurons. When this happens, astrocytes cannot produce enough lactate as an energy source for neurons, and this disrupts healthy brain metabolism and harms synapses” [senior author] Andreasson said. Blocking production of kynurenine by blocking IDO1 restores the ability of astrocytes to nourish neurons with lactate.
Offloading Coding Drudgery
Amazon used its AI assistant Q to save 4,500 developer-years of work upgrading legacy Java applications. The productivity boost translated into $260M in annualized savings. AWS is using all that money saved to provide AI compute directly to China in a way that circumvents US law regarding advanced chips.
“Strawberry Fields Forever”
The boomeranging of AI-startup founders back to the big platforms through a series of questionable deals is a significant divergence from prior platform cycles. Typically in Silicon Valley, we see employees splitting off to chase ideas while their former, saurischian employers slowly lose share of market growth to the next best thing. For example, Amazon built the infrastructure that allowed ideas to quickly become products without a large amount of capex, which allowed the outgrowth of cloud software, mobile apps, streaming video, etc. by a wave of pioneers who left legacy software and hardware companies. The situation with AI is different as the infrastructure cost is high, and the LLMs themselves are the infrastructure layer upon which applications will be built. In this analogy, ChatGPT and Gemini are the “AWS” of AI, and breakaway startup founders are returning to those large platforms simply because they can’t compete on their own. There are good reasons why foundational technologies like LLMs should be controlled by a small number of very large companies (see Search Win-Win), so this current round of entrepreneurs might be correct to cut their losses and return to the mother ships. The big question is whether the application layer (the myriad future apps that will be built on top of LLMs) will remain independent, driven by an army of visionary entrepreneurs, or be owned by the mega platforms. In the case of the iPhone, Apple has done alright for itself, but the really interesting tech has been the trillions of dollars of new industries (e.g., Uber, Meta, Amazon) built on top of, or greatly expanded by, the mobile operating systems. When Facebook IPO’d in 2012, they had no native mobile apps and it was unknown whether they could monetize a mobile newsfeed; currently, it’s a $1T company that wouldn’t exist without that foundational infrastructure layer of mobile phones, the app store, cell towers, etc. Now, Meta is trying to be the infrastructure layer of AI itself with Llama. Effectively, the companies that now dominate the application layer (Google with search, Meta with social, Microsoft with enterprise, Amazon with commerce, and, to a lesser extent, Apple with its App store) are also dominating the infrastructure layer for AI and may control its future application layer through walled gardens. Further, the same boomerang trend is playing out in the world of robotically embodied AI, with costs likely proving prohibitively high for independent operators as well. My base case is that we’ll see an explosion of new ideas similar to what we saw in the 2010s, but it’s not clear how easy it will be to pry them from the infrastructure-layer companies given their dominance in the application layer as well (which is where the vast wells of data reside to feed AI). With leading AI models about to leap forward in terms of reasoning (OpenAI’s Strawberry for example; also checkout NotebookLM from Google, it's a really impressive research assistant), I am optimistic we are on the precipice of a new wave of innovation that will make mobile and cloud look small in comparison. However, if entrepreneurs continue to be out-spent by the power-law platforms’ dominance, then the only logical preparation we can do is to rewatch Wall-E to get ready for our Buy-n-Large future.
Let me take you down
'Cause I'm going to strawberry fields
Nothing is real
And nothing to get hung about
Strawberry fields forever
Living is easy with eyes closed
Misunderstanding all you see
It's getting hard to be someone, but it all works out
-The Beatles
Miscellaneous Stuff
Neo vs. Cypher
Eighteen months ago, artist Nick Cave declared that ChatGPT’s lyrics in his style were a “grotesque mockery of what it is to be human”, which I discussed in a longer post about the importance of adapting to – rather than dismissing – new technology (#383). It’s become more clear over the last couple of years that the human brain functions like a LLM (and vice versa), and these forms of AI are capable of creativity. Recently, Cave was on The Reason Interview podcast and had this to say about AI:
If we're talking about music, the idea that music is a genuinely transformative, sort of transporting thing is being looked at with cynicism as well. We have like AI that has sort of song generating things, where you only have to put in a prompt, and a pretty good song pops out, right?
So, yeah, yeah. And, you know, it's as good as anything on the radio, and it's its first attempt, and in a year or two years time, we're going to be able to go straight to the product, and it's going to be indistinguishable about it between anything I can do, or Nirvana can do, or anybody else can Do, right? And this is, to me, an idea that that the creative struggle, which I think is the essence of meaning in this world, is seen as an impediment, or a kind of thing in the way to the product itself. Why? Why bother with having to sit down and kind of do soul searching and find out what sort of song you want to do or or go into a studio with your friends and try and create some sort of music. Why do we need that? When we have this product just drop out of this thing? And what scares me most of all is, I know I'm kind of ranting now, but whatever, what scares me most of all is that we are living in a society that is so demoralized that actually we don't really care. You know, there's a lot of people that say, Yeah, but we value true human art and performance and all that stuff. But I don't know, I think we can quite easily get to a place where no one cares, one way or the other, and so we're just losing these avenues for legitimate, transcendence.
Clearly, Cave has evolved his thinking concerning the capabilities of AI, but he seems more entrenched than ever in his argument that it’s not “human” and cannot represent what it is to be human. I think that is ultimately an indefensible position to take concerning the technology. What’s beautiful about LLMs is their computational irreducibility, as Stephen Wolfram explains:
The phenomenon of computational irreducibility leads to a fundamental tradeoff, of particular importance in thinking about things like AI. If we want to be able to know in advance—and broadly guarantee—what a system is going to do or be able to do, we have to set the system up to be computationally reducible. But if we want the system to be able to make the richest use of computation, it’ll inevitably be capable of computationally irreducible behavior. And it’s the same story with machine learning. If we want machine learning to be able to do the best it can, and perhaps give us the impression of “achieving magic”, then we have to allow it to show computational irreducibility. And if we want machine learning to be “understandable” it has to be computationally reducible, and not able to access the full power of computation.
At the outset, though, it’s not obvious whether machine learning actually has to access such power. It could be that there are computationally reducible ways to solve the kinds of problems we want to use machine learning to solve. But what we’ve discovered here is that even in solving very simple problems, the adaptive evolution process that’s at the heart of machine learning will end up sampling—and using—what we can expect to be computationally irreducible processes.
That excerpt comes from a post describing a mathematical effort by Wolfram to demonstrate the unexplainable methods of LLMs entitled: What’s Really Going On in Machine Learning?. Effectively, the process for adaptation in biological evolution can be analogized to machine learning. In consequence, it follows that the only reason to assume humans are special is a desire to ignore the truth. Rather than tracking in Cypher’s ill-fated footsteps and embracing seemingly blissful ignorance, I think a more useful (and perhaps existential) framework lies in harnessing technology as a means to our own, thoughtfully chosen ends because, in the not-too-distant future, very little is likely to be left in that uniquely human domain for which Cave so torturously longs.
Stuff About Demographics, the Economy, and Investing
Executive Blanding
CEOs have become more execution oriented and less creative since the GFC, according to a NBER working paper: “After the global financial crisis (GFC), the average interviewed CEO candidate has lower overall ability, is more execution oriented / less interpersonal, less charismatic and less creative/strategic than pre-GFC. Except for overall ability and execution oriented/interpersonal, these differences persist in hired CEOs. Interpersonal or ‘softer’ skills do not increase over time, either for CEO candidates or hired CEOs.” I tend to be skeptical of this sort of analysis given how easy it can be to find the answer you are looking for in a sea of data, but I have to agree that CEOs are not only getting more boring, but they increasingly seem to be falling victim to their own corporate narratives. One dichotomy of CEOs (explained in more detail in the book The Outsiders) can be defined by where they lie on the spectrum of capital allocation versus execution. We discussed this in more detail on page 17 of Complexity Investing, implying that CEOs that are more focused on capital allocation and decentralized decision making, and less on top-down execution, have a better shot at fostering adaptable organizations. My purely anecdotal feeling is that CEOs are more wrapped up today in the business news algorithm, by which I mean they are making far fewer independent decisions (i.e., exhibiting lower creativity), and are far more influenced by the reflexive narratives around their companies and, in many cases, the narrative around their own career prospects. It’s social networking’s algorithmic mind control applied to the hyperactive business news cycle. Boards also seem more eager to capitulate to activist investors, which, again, I think is characteristic of less independent thought and more narrative-driven decision making. Interestingly, CEO tenure overall is flat at around 8 years (it did dip down to ~7.5 during COVID before rebounding). My instinct would be that CEO turnover is up (at Starbucks, the CEOs last about as long as it takes to drink a venti mochaccino), but it seems the shift to more execution-oriented CEOs is not impacting turnover (perhaps less risk taking leads to less career risk). There is a large degree of variation by industry, according to the data from executive-compensation firm Equilar, with tech CEO tenure rising from ~7.5 to nearly 9 years since the GFC, while auto exec endurance fell from 9+ to 7 years. CEO turnover has been higher in consumer companies for the last few years as well. At NZS, we tend to look for companies where the CEO tries to make themself largely obsolete by building a powerful, decentralized organization, leaving execution to the biological organism that is the company’s many interacting employees.
✌️-Brad
Disclaimers:
The content of this newsletter is my personal opinion as of the date published and is subject to change without notice and may not reflect the opinion of NZS Capital, LLC. This newsletter is an informal gathering of topics I’ve recently read and thought about. I will sometimes state things in the newsletter that contradict my own views in order to provoke debate. Often I try to make jokes, and they aren’t very funny – sorry.
I may include links to third-party websites as a convenience, and the inclusion of such links does not imply any endorsement, approval, investigation, verification or monitoring by NZS Capital, LLC. If you choose to visit the linked sites, you do so at your own risk, and you will be subject to such sites' terms of use and privacy policies, over which NZS Capital, LLC has no control. In no event will NZS Capital, LLC be responsible for any information or content within the linked sites or your use of the linked sites.
Nothing in this newsletter should be construed as investment advice. The information contained herein is only as current as of the date indicated and may be superseded by subsequent market events or for other reasons. There is no guarantee that the information supplied is accurate, complete, or timely. Past performance is not a guarantee of future results.
Investing involves risk, including the possible loss of principal and fluctuation of value. Nothing contained in this newsletter is an offer to sell or solicit any investment services or securities. Initial Public Offerings (IPOs) are highly speculative investments and may be subject to lower liquidity and greater volatility. Special risks associated with IPOs include limited operating history, unseasoned trading, high turnover and non-repeatable performance.