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: the surprisingly high prevalence of cosmetic CGI in film and how AI is poised to alter the way everyone appears; Walmart leaps ahead in drones; new AI will allow you to have conversations with your past self, and perhaps even your future self; as products become more digital, it requires an entire new lens on outsourcing and vertical integration; learning about your own brain and how it sees the world and processes information; and, much more below...

Stuff about Innovation and Technology
Cinematic De-Aging
Upwards of 80-85% of Hollywood productions are digitally altered to touch up appearances or make actors look younger, oftentimes as part of actors’ contracts. This stat was relayed by Matt Panousis, visual effects veteran and COO of Monsters Aliens Robots Zombies (MARZ), on a recent The Town Podcast from Puck. MARZ has a new generative AI tool called Vanity AI, which fully automates cosmetic de-aging amidst the backdrop of an overworked and understaffed Hollywood VFX industry. The tool allows you to pick what appearance aspect you want to alter (crow’s feet, etc.) and adjust the degree with sliders; then, generative AI automatically creates the desired look throughout the entire movie (no CGI!). Panousis is quite bullish on the potential to vastly expand the use of such background VFX (it’s already beyond the Uncanny Valley, unnoticeable to viewers), and he thinks the AI is advancing so fast it’s hard to predict more than two months in advance how far ahead the tools will be. Likely within a couple years, Panousis reckons, actors will be digitally inserted wholesale into movies, complete with AI-generated speech. In #374, I wrote about Disney’s FRAN de-aging technology, as well as the movie The Congress, which deals with the near-future reality of actors retiring and granting AI the rights to their persona. Given that the majority of current advancements in AI tools are in large language models (LLMS) and generative graphics, Hollywood will be at the center of a major confrontation between talent and technology very soon. 

Walmart’s Drones Race Ahead
Walmart is bullish on drones. The giant retailer has curiously pulled far ahead of rivals Amazon and Google in the nascent drone delivery wars. They are expanding from three to 37 trial hubs across seven states and see drones as a greener alternative for deliveries and returns. Walmart’s drones can currently carry 100,000 different products up to 10 lbs (4.5 kg) and soon will be increasing capacity to 15 lbs (6.8 kg). Their vision involves drone fleets recharged entirely by solar power and deployable in emergencies. Can you believe it was way back in 2013 (10 years!) when Bezos made the splashy announcement for “Prime Air” drone delivery on 60 Minutes? (Here’s the promo video from December 2013 on YouTube). At long last, Amazon finally began drone deliveries last month in two US cities.

Back to the Future with Chatbots
GPT Index is a tool that connects LLMs with novel data (i.e., separate from the training set), allowing you to have a conversation with specific pools of information. In one example, an engineer wrote a program to allow queries about the content of specified videos. Imagine applying a similar tool to an entire library of podcasts or all of YouTube. The CEO of Every wrote a program that allowed him to query his old journal entries, as well as the transcript of a therapy session. He reported that “it felt like the AI knew me better than I knew myself”. I’m most intrigued with the idea of using LLMs to have a conversation with corporate data (emails, papers, presentations, meeting notes, etc.), which would allow you to travel back in time to effectively have a conversation with yourself to understand previous beliefs about an investment decision. It could be a great way to assist current decision making (or post-mortem analysis of prior choices) and discover/avoid cognitive bias mistakes. And, given enough data, you could even have a speculative conversation with your future self (although this possibility seems more plausible once it’s cost efficient to train LLMs on your personal data, rather than just inputting it as part of the prompt). Chatbot time travel...that's what I want! 

Digitalization Complexifies Outsourcing
High quality and low cost are a rare combination. Typically, you have to sacrifice some level of quality to make adequate margins on a product in a competitive market. During the Industrial Age, it was common for mature industries to outsource component manufacturing to marketplace suppliers, with competition resulting in reduced costs. The more mature an industry, the more likely it was to shift to a horizontal, outsourcing model, allowing the company to instead focus on design, brand, and other factors to gain market share. In these analog industries, dominated by linear rates of change, it was typically safer, from a productivity and cost perspective, to outsource rather than stay vertically integrated. There were exceptions to the outsourcing trend, as some niche companies, like RV-maker Winnebago, focused on quality over cost. When I toured a Winnebago factory at their former headquarters in Forest City, Iowa twenty years ago, I remember being surprised by how many components the company made themselves. At the time, they bought the large chassis or smaller vans from Detroit, but they made their own upholstery, cabinets, panels, molded plastic, etc. It felt like TVs were about the only thing they weren’t cranking out (although I am sure they still had a large supply network for other parts). Back then, only a few years out of college, I thought comparative advantage and globalization meant that companies always outsourced things they didn’t have an advantage in making. With the onset of the Digital Age, however, the question of outsourcing has become much thornier and more complex, leading to a resurgence of the Winnebago model of vertically integrated manufacturing. 

Throughout the analog-to-digital transition, companies have continued to (rather blindly) follow the outsourcing business model, with increasingly sophisticated components manufactured halfway around the world. However, there are overlooked, inherent dangers in outsourcing of complex, critical components as the economy transitions from analog to digital. As products rise in complexity and sophistication (e.g., integrating software, hardware, mobile, connectivity, and AI capabilities), more components have to interact with tighter tolerances to achieve functionality. Additionally, the half-lives of digital products are constantly being shortened, requiring accelerated innovation. These factors place renewed importance on both quality and adaptability, which are generally best achieved with an in-house, integrated testing and manufacturing process, where the company has the control and expertise to achieve disruptive innovation. In the past, we’ve also suggested that vertical integration is more likely, and perhaps inevitable, in many information-based businesses because it's key to unlocking network effects. From #219: 
I think it can be stated more generally that platforms of any type that have a data or informational advantage and network effect tend to vertically integrate. Vertical integration might be a necessary enabler of increasing network effects. What does this vertical integration trend suggest regarding other businesses where information is beginning to impact legacy, Industrial-Age sectors like healthcare, finance, and energy?

Despite digitalization’s impact on the outsourcing decision tree, too many companies remain entrenched in the Industrial Age “outsource everything” mentality. A good example is the auto industry. For decades, auto manufactures experienced a GDP-like growth rate, leading to increased outsourcing to cut costs, which, in turn, has hamstrung their ability to innovate. As a result, they were blindsided by the transition of a car from just a vehicle to a computer on wheels. Suddenly software, integration, chips, data, batteries, etc. became critical to success, but most car makers either never had, or had lost, skills in these areas. It’s much harder to reverse course and start designing and making parts and software yourself after you gave them up for the sake of driving margins in a mature industry. Tesla is an obvious counter example, insourcing software, data, batteries, and other technologies to its advantage to gain market share.
 
But, there are pitfalls to vertical integration as well. Insourcing often requires significant upfront costs, and, in a worst-case scenario, you can get locked into an inferior solution while a competitor with a better process or more expertise takes advantage. AMD’s shift to an external foundry while Intel struggled with their own process problems is one such example. And, in periods of stagnant (or negative) growth, vertically integrated companies still have a large overhead to bankroll. For example, hardware sales of devices like PCs, laptops, phones, and tablets declined 10% in 2022 and are forecast to decline again by 5% in 2023, according to Gartner. While much of this drop has to do with excess buying during the pandemic (and thus should rebound in the future), it also seems fair to say that many of these product categories have hit a rather static maturity and could benefit from outsourcing relationships – if supply chain resiliency wasn’t such a concern (e.g., Dell has set a goal to lower reliance on China by 2024).

Apple is a paradigm for the complexity of outsourcing decisions. For example, sometimes vertical integration is a strategic imperative, such as Apple's creation of their own semiconductors (here, they insourced the design but not the manufacturing). However, Bloomberg recently reported that Apple was looking to make their own phone displays. There are likely some broader reasons for this move (including the company’s desire to reduce reliance on Samsung as a supplier), however, displays are a fairly commoditized component (low supplier margins) with several competitors. Overall, Apple’s supply chain (largely reliant on TSMC and Foxconn), is perhaps one of the largest and most complex in the world and has become a large vulnerability for the company, as we’ve discussed over the last several years. The FT published a detailed two-part series on Apple’s China challenges (part one, part two) outlining many complexities of what it takes to manufacture the company’s products. Just thinking about Apple’s supply chain and how they decide to insource or outsource causes me mental paralysis. 

Hopefully, this discussion sheds some light on the complexity of outsourcing. In general, the faster the pace of innovation and the more digital the product, it would seem that vertical integration can be of real benefit. However, there appears to be no one correct blueprint for the right mix of outsourcing and vertical integration. What is clear is that, as software, data, and AI push deeper into more industries, it’s a topic that deserves much more attention and thoughtfulness. And, while I've focused on physical outsourcing here, the rapid rise of chatbots and LLM platforms will reshape the conversation around outsourcing of information-based white-collar functions as well. There are now a variety of important, compounding decisions companies are facing over where to focus internally vs. seek outside suppliers, and the key will be adaptability.

ChatGPT Concerns Google 
Google’s DeepMind chief, Demis Hassabis, likens the coming AI revolution to electrification, “an ‘epoch-defining’ technology—like the harnessing of electricity—that will change the very fabric of human life”, in a recent Time Magazine profile. It certainly feels like a lightbulb moment when you start to see all of the potential, particularly for huge productivity increases for information-based jobs. AI grows more potent by learning from humans, which, as Demis points out, makes us guinea pigs to an unprecedented degree. Compared to OpenAI and other startups, Google has taken a more cautious approach to the latest advances in LLMs and generative AI (which are built with the basics of the transformer models that Google researchers discovered in 2017). We could speculate that Google, having happened upon the advancement of transformer models first, might also be the furthest along – and may have seen something to give them pause. Or, perhaps vetting their guardrails has left them behind the competition, and their cautionary warnings are an attempt to scare competitors into walking instead of running to market with new AI products. A NYT report that Google founders Larry and Sergey had returned to the company's campus last month – to discuss how Google should respond to potential rival ChatGPT – suggests that the search giant has concerns regarding LLMs, including copyrighted training sources, risks of racism, and regulatory issues. I don’t know whether Google is simply being cautious or they are actually behind the curve, but it certainly seems that these models will continue to be released into the wild, and they are likely to be the new platforms for the next set of products and services humans create (see last week’s Follow the Developers for more). 

Something else noted in the Time article is how DeepMind was able to significantly improve LLM training efficiency using a model called Chinchilla. As I wrote a couple of weeks ago, we should get to a point where these new platforms can train daily (as opposed to the current ~yearly frequency). To be commercially successful, it seems like chatbots will require dramatic breakthroughs in efficiency, in part facilitated by moving from general-purpose processors like GPUs to custom chips like Google’s TPU (Andreessen Horowitz posted a lengthy review of the importance of chips for LLMs and generative AI). When training efficiency reaches this point, we are likely to enter a world where every human and countless connected devices have their own personalized, uniquely trained chatbot (see AI Companions), which implies a market billions of times larger than if chatbots were to remain general purpose.

Miscellaneous Stuff
Educating Against BS
Finland ranks number one for resilience against misinformation, thanks to an national education system that teaches skills for identifying fake or misleading news and information. As part of the media literacy core curriculum, students learn about algorithmic bias and how easy it is to manipulate videos. If we are to have any hope against the rising onslaught of fake everything, we should probably be teaching these skills to everyone, starting at the preschool level! I’d take this education a step further and teach kids to realize that, for the most part, the stories our brain concocts – about us, other people, and the surrounding world – may have very little bearing on reality. It’s important to understand fake news, but it’s perhaps even more valuable to be cognizant of our own internal social media "newsfeed" that our brains are constantly testing out. As I wrote recently in #372:
If you follow enough stories, tell enough stories, and try to make connections between enough stories, eventually you might get a little better at identifying which stories have some chance of being true, or at least teasing out the bits that might be more firmly embedded in reality. Among other activities, that’s how I see the profession of investing. We tell stories when we buy stocks and assemble a portfolio, trying very hard to find objective threads of evidence we can feed into our narratives. Then we look really closely to see if the story is true or not for each investment, as well as whether or not the story that defines the portfolio in totality has a chance at being true. We try to examine where our stories are vulnerable, or overly precise, in their embedded predictions. Stories are the heart of our pre-mortem process. I’ve been known to inform prospective clients that I am telling them a story and that it’s their job to decide if it has a chance of being true. CEOs tell stories about their companies and cultures. Salespeople tell stories about their products and services. Customers tell stories about why they consume those products and services. Politicians tell stories about society today and in the future. Your view of your “self” and your place in the world is merely a long running narrative your brain tells you about your time on Earth so far, which itself is largely influenced by the stories other people tell about you. 

Neurodiverse Thinking
Related to the storytelling brain we all have to learn to live with, this New Yorker article discusses the potentially different ways in which we think – visually vs. verbally vs. spatially, or some combination thereof. There are people with aphantasia, who are unable to picture nearly anything in their mind’s eye. I am not sure it’s possible to pin down how ideas move from the unconscious mind to our awareness in the real world, but spending a little time getting familiar with how your own brain tends to see the world might yield some insights into how best to avoid the pitfalls of common biases and improve decision making.

✌️-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.