SITALWeek #433
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: powering grids during the eclipse; police are increasingly using robots and droids to keep officers safe; the continuity problem in generative AI multimedia; a $100B supercomputer and a one trillion transistor chip are the future of AI; discovering novel flavors with AI for proteins; the "problem" with the economy is that forecasters are asking the wrong people about the economy; and, much more below.
Stuff about Innovation and Technology
Energy Eclipse
On April 8th, a total solar eclipse will cut a path across the US causing diminished sunlight for a couple of hours and several minutes of no sun for impacted areas. As the eclipse tracks across Texas, utilities are planning to leverage batteries to make up for the solar energy deficit. As we noted back in Green Star State, Texas has become the unlikely champion of going green, with the majority of their 27% growth in energy demand over the last decade fueled by solar, wind, and, perhaps more importantly, grid-connected battery storage. In contrast, the US overall experienced a puzzling pause in energy consumption, flatlining from 2007 to 2022 at around 4T KWh per year. By 2022, we were using 20% less energy than predicted despite decades of strong economic growth, likely in large part due to energy efficiency initiatives (LED light bulbs, appliances, insulation, building codes, etc.). Texas’s green capacity additions could serve as a model for the rest of the US as energy demand appears poised to accelerate, thanks to the perfect storm of the fossil-fuel transition, power-hungry artificial intelligence, and reshoring efforts (to name a few of the key factors). CBRE reported that data center construction grew 46% y/y in the second half of 2023 with 3078 megawatts of capacity added. Microsoft and OpenAI are planning a $100B AI supercomputer codenamed Stargate that would require several gigawatts of power. With multiple system-wide issues facing the grid, batteries combined with green energy sources could play a major role in stabilization. As EVs struggle to achieve mass market adoption in the US (and, finally, auto makers are pivoting to the more logical and greener PHEVs), I suspect many of the batteries targeted for new US EVs may end up as grid-connected power buffers. And, even nuclear power is making a comeback. Just last week, Michigan received a loan from the US government to reopen a closed nuclear plant. The site is also looking to add two small modular reactors down the line. I dove into more energy related topics recently in Pushing Electrons. Anyway, if you haven’t had a chance to see a total solar eclipse, I highly recommend it. You’ll want to have eclipse viewing glasses, binoculars (warning: only for use during complete totality – be very careful to avoid using them if even the smallest bit of sun is visible), and a light jacket – when the sun disappears entirely for a few minutes, it’s rather chilly.
Spot Shot
A Boston Dynamics Spot robot police dog named Roscoe was shot multiple times in the line of duty, able to stand back up and continue to approach a suspect until more shots disabled its communications with remote operators. The police credit the dog with potentially saving officers' lives, saying: “The incident provided a stark example of the benefits of mobile platforms capable of opening doors and ascending stairs in tactical missions involving armed suspects”. In related news, Boston Dynamics is set to reveal an updated version of Spot with far greater maneuverability and speed.
RealPage’s Reality Check
Following a NYT report that GM and other car makers were selling customers’ driving data – often without their knowledge – to insurance companies that used it to raise rates, GM has ended the practice. I discussed the rise in this type of algorithmic espionage in #431, and, in an early example from 2022, I covered the collusive rental pricing software RealPage in Algorithmic Distortion of Apartment Rents. Last week, the Department of Justice reportedly opened a criminal probe (in addition to its civil investigation) into RealPage and apartment owners.
Generative Incoherence
When you watch the batch of artist-created videos from OpenAI’s SORA engine, one thing that stands out is the lack of continuity. For example, even if you give the same language prompts to describe a character across multiple scenes, each instance will be rendered differently. This variation is evident in the video “Air Head”, about a man that has a yellow balloon for a head, with the balloon changing shape and often color tone from scene to scene. This issue is currently a major hurdle to adoption of generative AI tools for game development and media. In short, there needs to be a way to create reference characters, settings, etc. for use throughout worldbuilding to ensure consistency. Long-term memory, which seems to be a focal point for LLMs, should help, but might be technically tricky to apply in a complex, film-length creation.
1T GPU
For chip nerds, the chairman and chief scientist at TSMC discuss the path to 10x-ing to one trillion transistors to meet the insatiable computing demand of AI: “The computation and memory access required for AI training have increased by orders of magnitude in the past five years. Training GPT-3, for example, requires the equivalent of more than 5 billion billion operations per second of computation for an entire day (that’s 5,000 petaflops-days), and 3 trillion bytes (3 terabytes) of memory capacity. Both the computing power and the memory access needed for new generative AI applications continue to grow rapidly. We now need to answer a pressing question: How can semiconductor technology keep pace?” We discussed TSMC and the other pillars of the chip industry in our 2020 paper How a Handful of Chip Companies Came to Control the Fate of the World.
Miscellaneous Stuff
Encoding Sweetness with Amino Acids
After reading about FDA acceptance for the sugar substitute brazzein, a fruit-derived protein, I wondered if AlphaFold’s database and AI tools could lead to more protein-based sweetener agents or other flavor additives. I came across one ScienceDirect article from Korean researchers who modeled brazzein (which binds to the same taste receptors that sugar does) and believe AlphaFold could lead to more discoveries: We generated the brazzein and heterodimer complex model of taste receptors T1R2 and T1R3, for which both individual receptor structure models are now available in the AlphaFold Protein Structure Database. The docking analysis of the brazzein to T1R2/T1R3 heterocomplex may provide a useful structural basis to understand the flavour mechanism induced by sweet proteins.” I also wonder how much such techniques could be leveraged to amp up the addictive nature of food in the ongoing war between junk food makers and GLP-1 weight loss drugs. Perhaps there are entirely new flavors or combinations that nature didn't create that would dazzle our tastebuds.
Will AI Drug Discovery Bear Fruit?
Speaking of AlphaFold, the Economist discusses AI tools’ impact on drug development. While there are signs of hope, the range of outcomes remains wide open – myriad new treatments may be forthcoming, or we may discover that AI isn’t quite as proficient as we had hoped at picking the higher hanging fruit. One of the reasons prescription drug costs are so high is because it takes ~$6B in R&D to get the average drug to market, largely owing to the number of costly candidates that fail (MIT Technology Review). If AI can ultimately speed development and early selection of successful candidates, research costs should plummet for the industry, leading to much lower prices for consumers. In essence, the deflationary power of AI drug development could shrink the market while expanding the cures. So far, however, this promise remains elusive. The Technology Review article cites an estimated $18B invested in AI drug companies from 2012 to 2022 with scant reports of success to date. There is one recent Nature Biotechnology paper from Insilico that details how AI was used to get a drug to Phase II trials for the lung disease idiopathic pulmonary fibrosis. In other AI healthcare news, Google is making available HeAR, a model trained on audio recordings of millions of people for the purpose of diagnosing illnesses based on how a person breathes/coughs.
Stuff About Demographics, the Economy, and Investing
Survey Skew
A couple weeks back, at the end of #431, I mentioned a BI article that reported trending wage increases for lower (non-management) earners. It has been nagging at me that I didn’t talk about the more interesting element of that article, which is the increasingly unreliable survey data used in economic forecasting. Essentially, a number of factors are contributing to survey data being not only less reliable, but also skewed negative. Some of the reasons are interesting, such as the rise in spam calls causing a drop in answer rates, and some are more provocative, such as the personalities and demographics of people still willing to answer questions over the phone. The main takeaway is to be more skeptical of survey-based economic information. It’s rather surprising how much economic data is passed off as direct and statistically significant, when, in reality, it might just be a product of who picks up the phone. This factor appears particularly true of consumer confidence data. Economists always try to adjust for bias and say the data are still reliable, but this type of self-reporting appears increasingly non-representative of the economic whole. There are a lot of articles lately like this one in the WSJ that puzzle over why people are pessimistic when the economy is so strong, and there is a simple answer: only the people who picked up the phone are bummed out, while the majority of households are doing quite well. I would draw a contrast to this apparent near-term economic pessimism in the survey data with what might be a broader anxiety across the economy due to fear of AI and automation taking jobs (see Giving Up on the Old College Try). I went through several examples of the rapid pace at which AI is proving competent a couple weeks ago in You Are Special? So, perhaps people are genuinely concerned about the future, but, at the moment, that’s not stopping them from powering a strong economy here in the US.
✌️-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.